add test_data and test_strategy description (#4464)

* add test_data and test_strategy description

* update

docs/pipeline_usage/tutorials/ocr_pipelines/OCR.en.md

@@ -660,11 +660,11 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
 <td rowspan="7">OCR-default</td>
 <td>Intel 6271C</td>
-<td>3973.06</td>
+<td>3.97</td>
 <td>1015.40</td>
 <td>917.61</td>
 <td>4381.22</td>
@@ -676,7 +676,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3794.05</td>
+<td>3.79</td>
 <td>1022.50</td>
 <td>921.68</td>
 <td>4675.46</td>
@@ -688,7 +688,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>650.28</td>
+<td>0.65</td>
 <td>113.50</td>
 <td>102.48</td>
 <td>2240.15</td>
@@ -700,7 +700,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1062.29</td>
+<td>1.06</td>
 <td>114.90</td>
 <td>103.05</td>
 <td>2142.66</td>
@@ -712,7 +712,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>647.24</td>
+<td>0.65</td>
 <td>108.90</td>
 <td>101.95</td>
 <td>2456.05</td>
@@ -724,7 +724,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>735.63</td>
+<td>0.74</td>
 <td>115.90</td>
 <td>102.22</td>
 <td>2352.88</td>
@@ -736,7 +736,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1167.83</td>
+<td>1.17</td>
 <td>107.10</td>
 <td>101.78</td>
 <td>2361.88</td>
@@ -749,7 +749,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-mobile</td>
 <td>Intel 6271C</td>
-<td>1391.48</td>
+<td>1.39</td>
 <td>1019.60</td>
 <td>1007.69</td>
 <td>2178.12</td>
@@ -761,7 +761,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1151.22</td>
+<td>1.15</td>
 <td>1015.70</td>
 <td>1006.87</td>
 <td>2184.91</td>
@@ -773,7 +773,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>353.96</td>
+<td>0.35</td>
 <td>110.80</td>
 <td>103.77</td>
 <td>2022.49</td>
@@ -785,7 +785,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>266.79</td>
+<td>0.27</td>
 <td>110.90</td>
 <td>103.80</td>
 <td>1762.36</td>
@@ -797,7 +797,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>553.80</td>
+<td>0.55</td>
 <td>113.80</td>
 <td>103.68</td>
 <td>1728.02</td>
@@ -809,7 +809,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>215.95</td>
+<td>0.22</td>
 <td>111.90</td>
 <td>103.99</td>
 <td>2073.88</td>
@@ -821,7 +821,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>305.25</td>
+<td>0.31</td>
 <td>119.90</td>
 <td>104.24</td>
 <td>2037.38</td>
@@ -833,7 +833,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>6511.77</td>
+<td>6.51</td>
 <td>147.30</td>
 <td>106.24</td>
 <td>3550.58</td>
@@ -845,7 +845,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>456.25</td>
+<td>0.46</td>
 <td>111.90</td>
 <td>103.11</td>
 <td>2035.38</td>
@@ -858,7 +858,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="7">OCR-nopp-server</td>
 <td>Intel 6271C</td>
-<td>2996.17</td>
+<td>3.00</td>
 <td>1016.00</td>
 <td>1004.87</td>
 <td>4445.46</td>
@@ -870,7 +870,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3233.52</td>
+<td>3.23</td>
 <td>1010.70</td>
 <td>1002.63</td>
 <td>4175.39</td>
@@ -882,7 +882,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>337.05</td>
+<td>0.34</td>
 <td>110.90</td>
 <td>103.30</td>
 <td>1904.99</td>
@@ -894,7 +894,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>692.43</td>
+<td>0.69</td>
 <td>108.90</td>
 <td>102.95</td>
 <td>1808.30</td>
@@ -906,7 +906,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>376.60</td>
+<td>0.38</td>
 <td>109.40</td>
 <td>102.34</td>
 <td>2100.00</td>
@@ -918,7 +918,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>412.03</td>
+<td>0.41</td>
 <td>109.00</td>
 <td>103.18</td>
 <td>2055.21</td>
@@ -930,7 +930,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>822.98</td>
+<td>0.82</td>
 <td>104.40</td>
 <td>101.73</td>
 <td>1906.88</td>
@@ -943,7 +943,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-min736-mobile</td>
 <td>Intel 6271C</td>
-<td>1406.44</td>
+<td>1.41</td>
 <td>1020.10</td>
 <td>1008.14</td>
 <td>2184.16</td>
@@ -955,7 +955,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1204.44</td>
+<td>1.20</td>
 <td>1015.70</td>
 <td>1007.08</td>
 <td>2254.04</td>
@@ -967,7 +967,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>364.67</td>
+<td>0.36</td>
 <td>112.90</td>
 <td>104.29</td>
 <td>2174.58</td>
@@ -979,7 +979,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>267.13</td>
+<td>0.27</td>
 <td>113.90</td>
 <td>104.48</td>
 <td>1717.55</td>
@@ -991,7 +991,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>573.26</td>
+<td>0.57</td>
 <td>118.80</td>
 <td>104.45</td>
 <td>1693.10</td>
@@ -1003,7 +1003,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>223.77</td>
+<td>0.22</td>
 <td>113.40</td>
 <td>104.66</td>
 <td>2037.13</td>
@@ -1015,7 +1015,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>310.30</td>
+<td>0.31</td>
 <td>119.30</td>
 <td>106.05</td>
 <td>1879.15</td>
@@ -1027,7 +1027,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>6389.91</td>
+<td>6.39</td>
 <td>124.90</td>
 <td>107.16</td>
 <td>3578.98</td>
@@ -1039,7 +1039,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>474.07</td>
+<td>0.47</td>
 <td>109.60</td>
 <td>103.26</td>
 <td>1961.40</td>
@@ -1052,7 +1052,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="7">OCR-nopp-min736-server</td>
 <td>Intel 6271C</td>
-<td>3255.36</td>
+<td>3.26</td>
 <td>1068.50</td>
 <td>1004.96</td>
 <td>4582.52</td>
@@ -1064,7 +1064,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3516.01</td>
+<td>3.52</td>
 <td>1010.70</td>
 <td>1002.33</td>
 <td>4723.23</td>
@@ -1076,7 +1076,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>345.96</td>
+<td>0.35</td>
 <td>108.90</td>
 <td>103.94</td>
 <td>1703.65</td>
@@ -1088,7 +1088,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>710.22</td>
+<td>0.71</td>
 <td>110.80</td>
 <td>103.54</td>
 <td>1800.06</td>
@@ -1100,7 +1100,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>401.64</td>
+<td>0.40</td>
 <td>110.20</td>
 <td>102.75</td>
 <td>2012.64</td>
@@ -1112,7 +1112,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>440.96</td>
+<td>0.44</td>
 <td>114.90</td>
 <td>103.87</td>
 <td>2002.72</td>
@@ -1124,7 +1124,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>885.37</td>
+<td>0.89</td>
 <td>105.00</td>
 <td>101.91</td>
 <td>2149.31</td>
@@ -1137,7 +1137,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-max640-mobile</td>
 <td>Intel 6271C</td>
-<td>1003.18</td>
+<td>1.00</td>
 <td>1033.70</td>
 <td>1005.95</td>
 <td>2021.88</td>
@@ -1149,7 +1149,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>880.14</td>
+<td>0.88</td>
 <td>1043.60</td>
 <td>1006.77</td>
 <td>1980.82</td>
@@ -1161,7 +1161,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>282.01</td>
+<td>0.28</td>
 <td>125.70</td>
 <td>101.56</td>
 <td>1962.27</td>
@@ -1173,7 +1173,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>213.70</td>
+<td>0.21</td>
 <td>122.50</td>
 <td>101.87</td>
 <td>1772.39</td>
@@ -1185,7 +1185,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>428.69</td>
+<td>0.43</td>
 <td>133.80</td>
 <td>101.82</td>
 <td>1636.93</td>
@@ -1197,7 +1197,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>175.62</td>
+<td>0.18</td>
 <td>119.90</td>
 <td>102.12</td>
 <td>2119.93</td>
@@ -1209,7 +1209,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>243.30</td>
+<td>0.24</td>
 <td>126.80</td>
 <td>101.78</td>
 <td>1905.14</td>
@@ -1221,7 +1221,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>7084.63</td>
+<td>7.08</td>
 <td>137.80</td>
 <td>104.83</td>
 <td>2931.08</td>
@@ -1233,7 +1233,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>364.92</td>
+<td>0.36</td>
 <td>124.80</td>
 <td>101.70</td>
 <td>1983.21</td>
@@ -1246,7 +1246,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-max960-mobile</td>
 <td>Intel 6271C</td>
-<td>1207.47</td>
+<td>1.21</td>
 <td>1020.00</td>
 <td>1008.49</td>
 <td>2200.30</td>
@@ -1258,7 +1258,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1005.80</td>
+<td>1.01</td>
 <td>1024.10</td>
 <td>1007.32</td>
 <td>2038.80</td>
@@ -1270,7 +1270,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>315.36</td>
+<td>0.32</td>
 <td>107.50</td>
 <td>102.00</td>
 <td>2001.21</td>
@@ -1282,7 +1282,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>232.34</td>
+<td>0.23</td>
 <td>107.70</td>
 <td>102.33</td>
 <td>1727.89</td>
@@ -1294,7 +1294,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>486.29</td>
+<td>0.49</td>
 <td>109.90</td>
 <td>102.26</td>
 <td>1726.01</td>
@@ -1306,7 +1306,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>199.00</td>
+<td>0.20</td>
 <td>109.90</td>
 <td>102.52</td>
 <td>1959.46</td>
@@ -1318,7 +1318,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>270.46</td>
+<td>0.27</td>
 <td>102.90</td>
 <td>101.19</td>
 <td>1938.48</td>
@@ -1330,7 +1330,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>5441.59</td>
+<td>5.44</td>
 <td>122.10</td>
 <td>105.91</td>
 <td>3094.72</td>
@@ -1342,7 +1342,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>413.04</td>
+<td>0.41</td>
 <td>106.00</td>
 <td>101.81</td>
 <td>1859.88</td>
@@ -1355,7 +1355,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-max640-server</td>
 <td>Intel 6271C</td>
-<td>2163.69</td>
+<td>2.16</td>
 <td>1026.30</td>
 <td>1005.10</td>
 <td>3467.93</td>
@@ -1367,7 +1367,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>2301.92</td>
+<td>2.30</td>
 <td>1008.70</td>
 <td>1003.32</td>
 <td>3435.54</td>
@@ -1379,7 +1379,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>350.11</td>
+<td>0.35</td>
 <td>104.70</td>
 <td>101.27</td>
 <td>1948.85</td>
@@ -1391,7 +1391,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>252.81</td>
+<td>0.25</td>
 <td>104.90</td>
 <td>101.42</td>
 <td>1833.93</td>
@@ -1403,7 +1403,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>555.28</td>
+<td>0.56</td>
 <td>106.20</td>
 <td>101.47</td>
 <td>1669.73</td>
@@ -1415,7 +1415,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>233.91</td>
+<td>0.23</td>
 <td>109.40</td>
 <td>101.45</td>
 <td>1968.77</td>
@@ -1427,7 +1427,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>302.47</td>
+<td>0.30</td>
 <td>106.10</td>
 <td>101.55</td>
 <td>2027.13</td>
@@ -1439,7 +1439,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>7263.68</td>
+<td>7.26</td>
 <td>133.90</td>
 <td>104.48</td>
 <td>5473.38</td>
@@ -1451,7 +1451,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>576.16</td>
+<td>0.58</td>
 <td>103.90</td>
 <td>100.86</td>
 <td>1884.23</td>
@@ -1464,7 +1464,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-max960-server</td>
 <td>Intel 6271C</td>
-<td>2531.42</td>
+<td>2.53</td>
 <td>1014.50</td>
 <td>1005.22</td>
 <td>3625.57</td>
@@ -1476,7 +1476,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>2661.74</td>
+<td>2.66</td>
 <td>1010.60</td>
 <td>1003.39</td>
 <td>3580.64</td>
@@ -1488,7 +1488,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>399.72</td>
+<td>0.40</td>
 <td>105.90</td>
 <td>101.76</td>
 <td>2040.65</td>
@@ -1500,7 +1500,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>289.63</td>
+<td>0.29</td>
 <td>108.90</td>
 <td>102.12</td>
 <td>1821.03</td>
@@ -1512,7 +1512,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>601.65</td>
+<td>0.60</td>
 <td>109.90</td>
 <td>101.98</td>
 <td>1797.75</td>
@@ -1524,7 +1524,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>281.77</td>
+<td>0.28</td>
 <td>108.80</td>
 <td>101.92</td>
 <td>2016.22</td>
@@ -1536,7 +1536,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>342.18</td>
+<td>0.34</td>
 <td>111.00</td>
 <td>102.75</td>
 <td>1964.21</td>
@@ -1548,7 +1548,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>6277.42</td>
+<td>6.28</td>
 <td>129.10</td>
 <td>103.74</td>
 <td>7780.70</td>
@@ -1560,7 +1560,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>672.28</td>
+<td>0.67</td>
 <td>116.90</td>
 <td>101.33</td>
 <td>1941.09</td>
@@ -1573,7 +1573,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="7">OCR-nopp-min1280-server</td>
 <td>Intel 6271C</td>
-<td>4128.58</td>
+<td>4.13</td>
 <td>1043.40</td>
 <td>1005.45</td>
 <td>5993.70</td>
@@ -1585,7 +1585,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>4464.14</td>
+<td>4.46</td>
 <td>1011.70</td>
 <td>996.72</td>
 <td>5633.51</td>
@@ -1597,7 +1597,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>418.86</td>
+<td>0.42</td>
 <td>113.90</td>
 <td>106.08</td>
 <td>1747.88</td>
@@ -1609,7 +1609,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>816.32</td>
+<td>0.82</td>
 <td>116.80</td>
 <td>105.18</td>
 <td>1873.38</td>
@@ -1621,7 +1621,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>554.14</td>
+<td>0.55</td>
 <td>114.80</td>
 <td>103.14</td>
 <td>2036.36</td>
@@ -1633,7 +1633,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>547.42</td>
+<td>0.55</td>
 <td>105.90</td>
 <td>101.86</td>
 <td>1931.35</td>
@@ -1645,7 +1645,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1132.56</td>
+<td>1.13</td>
 <td>105.90</td>
 <td>102.35</td>
 <td>2066.73</td>
@@ -1658,7 +1658,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">OCR-nopp-min1280-mobile</td>
 <td>Intel 6271C</td>
-<td>1586.51</td>
+<td>1.59</td>
 <td>1019.90</td>
 <td>1008.39</td>
 <td>2366.86</td>
@@ -1670,7 +1670,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1292.65</td>
+<td>1.29</td>
 <td>1017.70</td>
 <td>1007.28</td>
 <td>2501.24</td>
@@ -1682,7 +1682,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>426.04</td>
+<td>0.43</td>
 <td>120.90</td>
 <td>107.02</td>
 <td>2108.87</td>
@@ -1694,7 +1694,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>288.54</td>
+<td>0.29</td>
 <td>117.90</td>
 <td>107.19</td>
 <td>1847.97</td>
@@ -1706,7 +1706,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>610.11</td>
+<td>0.61</td>
 <td>122.80</td>
 <td>107.07</td>
 <td>1789.25</td>
@@ -1718,7 +1718,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>239.82</td>
+<td>0.24</td>
 <td>116.80</td>
 <td>106.80</td>
 <td>2092.63</td>
@@ -1730,7 +1730,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>340.15</td>
+<td>0.34</td>
 <td>125.80</td>
 <td>106.79</td>
 <td>1959.45</td>
@@ -1742,7 +1742,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>6636.03</td>
+<td>6.64</td>
 <td>139.40</td>
 <td>107.63</td>
 <td>4283.97</td>
@@ -1754,7 +1754,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>511.86</td>
+<td>0.51</td>
 <td>116.90</td>
 <td>105.06</td>
 <td>1927.22</td>
@@ -1821,7 +1821,10 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing 200 images from document and general scenarios.
+* Test strategy:
+    * Warm up with 20 samples, then repeat the full dataset once for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/ocr_pipelines/OCR.md

@@ -652,11 +652,11 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="7">OCR-default</td>
 <td>Intel 6271C</td>
-<td>3973.06</td>
+<td>3.97</td>
 <td>1015.40</td>
 <td>917.61</td>
 <td>4381.22</td>
@@ -668,7 +668,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3794.05</td>
+<td>3.79</td>
 <td>1022.50</td>
 <td>921.68</td>
 <td>4675.46</td>
@@ -680,7 +680,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>650.28</td>
+<td>0.65</td>
 <td>113.50</td>
 <td>102.48</td>
 <td>2240.15</td>
@@ -692,7 +692,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1062.29</td>
+<td>1.06</td>
 <td>114.90</td>
 <td>103.05</td>
 <td>2142.66</td>
@@ -704,7 +704,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>647.24</td>
+<td>0.65</td>
 <td>108.90</td>
 <td>101.95</td>
 <td>2456.05</td>
@@ -716,7 +716,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>735.63</td>
+<td>0.74</td>
 <td>115.90</td>
 <td>102.22</td>
 <td>2352.88</td>
@@ -728,7 +728,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1167.83</td>
+<td>1.17</td>
 <td>107.10</td>
 <td>101.78</td>
 <td>2361.88</td>
@@ -741,7 +741,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-mobile</td>
 <td>Intel 6271C</td>
-<td>1391.48</td>
+<td>1.39</td>
 <td>1019.60</td>
 <td>1007.69</td>
 <td>2178.12</td>
@@ -753,7 +753,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1151.22</td>
+<td>1.15</td>
 <td>1015.70</td>
 <td>1006.87</td>
 <td>2184.91</td>
@@ -765,7 +765,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>353.96</td>
+<td>0.35</td>
 <td>110.80</td>
 <td>103.77</td>
 <td>2022.49</td>
@@ -777,7 +777,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>266.79</td>
+<td>0.27</td>
 <td>110.90</td>
 <td>103.80</td>
 <td>1762.36</td>
@@ -789,7 +789,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>553.80</td>
+<td>0.55</td>
 <td>113.80</td>
 <td>103.68</td>
 <td>1728.02</td>
@@ -801,7 +801,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>215.95</td>
+<td>0.22</td>
 <td>111.90</td>
 <td>103.99</td>
 <td>2073.88</td>
@@ -813,7 +813,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>305.25</td>
+<td>0.31</td>
 <td>119.90</td>
 <td>104.24</td>
 <td>2037.38</td>
@@ -825,7 +825,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>6511.77</td>
+<td>6.51</td>
 <td>147.30</td>
 <td>106.24</td>
 <td>3550.58</td>
@@ -837,7 +837,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>456.25</td>
+<td>0.46</td>
 <td>111.90</td>
 <td>103.11</td>
 <td>2035.38</td>
@@ -850,7 +850,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="7">OCR-nopp-server</td>
 <td>Intel 6271C</td>
-<td>2996.17</td>
+<td>3.00</td>
 <td>1016.00</td>
 <td>1004.87</td>
 <td>4445.46</td>
@@ -862,7 +862,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3233.52</td>
+<td>3.23</td>
 <td>1010.70</td>
 <td>1002.63</td>
 <td>4175.39</td>
@@ -874,7 +874,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>337.05</td>
+<td>0.34</td>
 <td>110.90</td>
 <td>103.30</td>
 <td>1904.99</td>
@@ -886,7 +886,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>692.43</td>
+<td>0.69</td>
 <td>108.90</td>
 <td>102.95</td>
 <td>1808.30</td>
@@ -898,7 +898,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>376.60</td>
+<td>0.38</td>
 <td>109.40</td>
 <td>102.34</td>
 <td>2100.00</td>
@@ -910,7 +910,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>412.03</td>
+<td>0.41</td>
 <td>109.00</td>
 <td>103.18</td>
 <td>2055.21</td>
@@ -922,7 +922,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>822.98</td>
+<td>0.82</td>
 <td>104.40</td>
 <td>101.73</td>
 <td>1906.88</td>
@@ -935,7 +935,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-min736-mobile</td>
 <td>Intel 6271C</td>
-<td>1406.44</td>
+<td>1.41</td>
 <td>1020.10</td>
 <td>1008.14</td>
 <td>2184.16</td>
@@ -947,7 +947,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1204.44</td>
+<td>1.20</td>
 <td>1015.70</td>
 <td>1007.08</td>
 <td>2254.04</td>
@@ -959,7 +959,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>364.67</td>
+<td>0.36</td>
 <td>112.90</td>
 <td>104.29</td>
 <td>2174.58</td>
@@ -971,7 +971,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>267.13</td>
+<td>0.27</td>
 <td>113.90</td>
 <td>104.48</td>
 <td>1717.55</td>
@@ -983,7 +983,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>573.26</td>
+<td>0.57</td>
 <td>118.80</td>
 <td>104.45</td>
 <td>1693.10</td>
@@ -995,7 +995,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>223.77</td>
+<td>0.22</td>
 <td>113.40</td>
 <td>104.66</td>
 <td>2037.13</td>
@@ -1007,7 +1007,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>310.30</td>
+<td>0.31</td>
 <td>119.30</td>
 <td>106.05</td>
 <td>1879.15</td>
@@ -1019,7 +1019,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>6389.91</td>
+<td>6.39</td>
 <td>124.90</td>
 <td>107.16</td>
 <td>3578.98</td>
@@ -1031,7 +1031,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>474.07</td>
+<td>0.47</td>
 <td>109.60</td>
 <td>103.26</td>
 <td>1961.40</td>
@@ -1044,7 +1044,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="7">OCR-nopp-min736-server</td>
 <td>Intel 6271C</td>
-<td>3255.36</td>
+<td>3.26</td>
 <td>1068.50</td>
 <td>1004.96</td>
 <td>4582.52</td>
@@ -1056,7 +1056,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3516.01</td>
+<td>3.52</td>
 <td>1010.70</td>
 <td>1002.33</td>
 <td>4723.23</td>
@@ -1068,7 +1068,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>345.96</td>
+<td>0.35</td>
 <td>108.90</td>
 <td>103.94</td>
 <td>1703.65</td>
@@ -1080,7 +1080,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>710.22</td>
+<td>0.71</td>
 <td>110.80</td>
 <td>103.54</td>
 <td>1800.06</td>
@@ -1092,7 +1092,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>401.64</td>
+<td>0.40</td>
 <td>110.20</td>
 <td>102.75</td>
 <td>2012.64</td>
@@ -1104,7 +1104,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>440.96</td>
+<td>0.44</td>
 <td>114.90</td>
 <td>103.87</td>
 <td>2002.72</td>
@@ -1116,7 +1116,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>885.37</td>
+<td>0.89</td>
 <td>105.00</td>
 <td>101.91</td>
 <td>2149.31</td>
@@ -1129,7 +1129,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-max640-mobile</td>
 <td>Intel 6271C</td>
-<td>1003.18</td>
+<td>1.00</td>
 <td>1033.70</td>
 <td>1005.95</td>
 <td>2021.88</td>
@@ -1141,7 +1141,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>880.14</td>
+<td>0.88</td>
 <td>1043.60</td>
 <td>1006.77</td>
 <td>1980.82</td>
@@ -1153,7 +1153,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>282.01</td>
+<td>0.28</td>
 <td>125.70</td>
 <td>101.56</td>
 <td>1962.27</td>
@@ -1165,7 +1165,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>213.70</td>
+<td>0.21</td>
 <td>122.50</td>
 <td>101.87</td>
 <td>1772.39</td>
@@ -1177,7 +1177,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>428.69</td>
+<td>0.43</td>
 <td>133.80</td>
 <td>101.82</td>
 <td>1636.93</td>
@@ -1189,7 +1189,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>175.62</td>
+<td>0.18</td>
 <td>119.90</td>
 <td>102.12</td>
 <td>2119.93</td>
@@ -1201,7 +1201,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>243.30</td>
+<td>0.24</td>
 <td>126.80</td>
 <td>101.78</td>
 <td>1905.14</td>
@@ -1213,7 +1213,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>7084.63</td>
+<td>7.08</td>
 <td>137.80</td>
 <td>104.83</td>
 <td>2931.08</td>
@@ -1225,7 +1225,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>364.92</td>
+<td>0.36</td>
 <td>124.80</td>
 <td>101.70</td>
 <td>1983.21</td>
@@ -1238,7 +1238,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-max960-mobile</td>
 <td>Intel 6271C</td>
-<td>1207.47</td>
+<td>1.21</td>
 <td>1020.00</td>
 <td>1008.49</td>
 <td>2200.30</td>
@@ -1250,7 +1250,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1005.80</td>
+<td>1.01</td>
 <td>1024.10</td>
 <td>1007.32</td>
 <td>2038.80</td>
@@ -1262,7 +1262,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>315.36</td>
+<td>0.32</td>
 <td>107.50</td>
 <td>102.00</td>
 <td>2001.21</td>
@@ -1274,7 +1274,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>232.34</td>
+<td>0.23</td>
 <td>107.70</td>
 <td>102.33</td>
 <td>1727.89</td>
@@ -1286,7 +1286,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>486.29</td>
+<td>0.49</td>
 <td>109.90</td>
 <td>102.26</td>
 <td>1726.01</td>
@@ -1298,7 +1298,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>199.00</td>
+<td>0.20</td>
 <td>109.90</td>
 <td>102.52</td>
 <td>1959.46</td>
@@ -1310,7 +1310,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>270.46</td>
+<td>0.27</td>
 <td>102.90</td>
 <td>101.19</td>
 <td>1938.48</td>
@@ -1322,7 +1322,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>5441.59</td>
+<td>5.44</td>
 <td>122.10</td>
 <td>105.91</td>
 <td>3094.72</td>
@@ -1334,7 +1334,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>413.04</td>
+<td>0.41</td>
 <td>106.00</td>
 <td>101.81</td>
 <td>1859.88</td>
@@ -1347,7 +1347,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-max640-server</td>
 <td>Intel 6271C</td>
-<td>2163.69</td>
+<td>2.16</td>
 <td>1026.30</td>
 <td>1005.10</td>
 <td>3467.93</td>
@@ -1359,7 +1359,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>2301.92</td>
+<td>2.30</td>
 <td>1008.70</td>
 <td>1003.32</td>
 <td>3435.54</td>
@@ -1371,7 +1371,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>350.11</td>
+<td>0.35</td>
 <td>104.70</td>
 <td>101.27</td>
 <td>1948.85</td>
@@ -1383,7 +1383,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>252.81</td>
+<td>0.25</td>
 <td>104.90</td>
 <td>101.42</td>
 <td>1833.93</td>
@@ -1395,7 +1395,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>555.28</td>
+<td>0.56</td>
 <td>106.20</td>
 <td>101.47</td>
 <td>1669.73</td>
@@ -1407,7 +1407,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>233.91</td>
+<td>0.23</td>
 <td>109.40</td>
 <td>101.45</td>
 <td>1968.77</td>
@@ -1419,7 +1419,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>302.47</td>
+<td>0.30</td>
 <td>106.10</td>
 <td>101.55</td>
 <td>2027.13</td>
@@ -1431,7 +1431,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>7263.68</td>
+<td>7.26</td>
 <td>133.90</td>
 <td>104.48</td>
 <td>5473.38</td>
@@ -1443,7 +1443,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>576.16</td>
+<td>0.58</td>
 <td>103.90</td>
 <td>100.86</td>
 <td>1884.23</td>
@@ -1456,7 +1456,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-max960-server</td>
 <td>Intel 6271C</td>
-<td>2531.42</td>
+<td>2.53</td>
 <td>1014.50</td>
 <td>1005.22</td>
 <td>3625.57</td>
@@ -1468,7 +1468,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>2661.74</td>
+<td>2.66</td>
 <td>1010.60</td>
 <td>1003.39</td>
 <td>3580.64</td>
@@ -1480,7 +1480,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>399.72</td>
+<td>0.40</td>
 <td>105.90</td>
 <td>101.76</td>
 <td>2040.65</td>
@@ -1492,7 +1492,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>289.63</td>
+<td>0.29</td>
 <td>108.90</td>
 <td>102.12</td>
 <td>1821.03</td>
@@ -1504,7 +1504,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>601.65</td>
+<td>0.60</td>
 <td>109.90</td>
 <td>101.98</td>
 <td>1797.75</td>
@@ -1516,7 +1516,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>281.77</td>
+<td>0.28</td>
 <td>108.80</td>
 <td>101.92</td>
 <td>2016.22</td>
@@ -1528,7 +1528,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>342.18</td>
+<td>0.34</td>
 <td>111.00</td>
 <td>102.75</td>
 <td>1964.21</td>
@@ -1540,7 +1540,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>6277.42</td>
+<td>6.28</td>
 <td>129.10</td>
 <td>103.74</td>
 <td>7780.70</td>
@@ -1552,7 +1552,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>672.28</td>
+<td>0.67</td>
 <td>116.90</td>
 <td>101.33</td>
 <td>1941.09</td>
@@ -1565,7 +1565,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="7">OCR-nopp-min1280-server</td>
 <td>Intel 6271C</td>
-<td>4128.58</td>
+<td>4.13</td>
 <td>1043.40</td>
 <td>1005.45</td>
 <td>5993.70</td>
@@ -1577,7 +1577,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>4464.14</td>
+<td>4.46</td>
 <td>1011.70</td>
 <td>996.72</td>
 <td>5633.51</td>
@@ -1589,7 +1589,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>418.86</td>
+<td>0.42</td>
 <td>113.90</td>
 <td>106.08</td>
 <td>1747.88</td>
@@ -1601,7 +1601,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>816.32</td>
+<td>0.82</td>
 <td>116.80</td>
 <td>105.18</td>
 <td>1873.38</td>
@@ -1613,7 +1613,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>554.14</td>
+<td>0.55</td>
 <td>114.80</td>
 <td>103.14</td>
 <td>2036.36</td>
@@ -1625,7 +1625,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>547.42</td>
+<td>0.55</td>
 <td>105.90</td>
 <td>101.86</td>
 <td>1931.35</td>
@@ -1637,7 +1637,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1132.56</td>
+<td>1.13</td>
 <td>105.90</td>
 <td>102.35</td>
 <td>2066.73</td>
@@ -1650,7 +1650,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">OCR-nopp-min1280-mobile</td>
 <td>Intel 6271C</td>
-<td>1586.51</td>
+<td>1.59</td>
 <td>1019.90</td>
 <td>1008.39</td>
 <td>2366.86</td>
@@ -1662,7 +1662,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1292.65</td>
+<td>1.29</td>
 <td>1017.70</td>
 <td>1007.28</td>
 <td>2501.24</td>
@@ -1674,7 +1674,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>426.04</td>
+<td>0.43</td>
 <td>120.90</td>
 <td>107.02</td>
 <td>2108.87</td>
@@ -1686,7 +1686,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>288.54</td>
+<td>0.29</td>
 <td>117.90</td>
 <td>107.19</td>
 <td>1847.97</td>
@@ -1698,7 +1698,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>610.11</td>
+<td>0.61</td>
 <td>122.80</td>
 <td>107.07</td>
 <td>1789.25</td>
@@ -1710,7 +1710,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>239.82</td>
+<td>0.24</td>
 <td>116.80</td>
 <td>106.80</td>
 <td>2092.63</td>
@@ -1722,7 +1722,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>340.15</td>
+<td>0.34</td>
 <td>125.80</td>
 <td>106.79</td>
 <td>1959.45</td>
@@ -1734,7 +1734,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>6636.03</td>
+<td>6.64</td>
 <td>139.40</td>
 <td>107.63</td>
 <td>4283.97</td>
@@ -1746,7 +1746,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>511.86</td>
+<td>0.51</td>
 <td>116.90</td>
 <td>105.06</td>
 <td>1927.22</td>
@@ -1813,7 +1813,10 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含文档场景和通用场景的200张图像。
+* 测试策略：
+    * 使用 20 个样本进行预热，然后对整个数据集重复 1 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
 

docs/pipeline_usage/tutorials/ocr_pipelines/PP-StructureV3.en.md

@@ -780,11 +780,11 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
 <td rowspan="5">PP_StructureV3-default</td>
 <td>Intel 8350C + A100</td>
-<td>1378.46</td>
+<td>1.38</td>
 <td>1384.60</td>
 <td>113.26</td>
 <td>5781.59</td>
@@ -796,7 +796,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2382.03</td>
+<td>2.38</td>
 <td>608.70</td>
 <td>109.96</td>
 <td>6388.91</td>
@@ -808,7 +808,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1364.42</td>
+<td>1.36</td>
 <td>744.30</td>
 <td>112.82</td>
 <td>6199.01</td>
@@ -820,7 +820,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1735.04</td>
+<td>1.74</td>
 <td>418.50</td>
 <td>105.96</td>
 <td>6138.25</td>
@@ -832,7 +832,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>3700.82</td>
+<td>3.70</td>
 <td>434.40</td>
 <td>105.45</td>
 <td>6865.87</td>
@@ -845,7 +845,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="3">PP_StructureV3-pp</td>
 <td>Intel 8350C + A100</td>
-<td>3498.87</td>
+<td>3.50</td>
 <td>679.30</td>
 <td>105.96</td>
 <td>13850.20</td>
@@ -857,7 +857,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>5030.33</td>
+<td>5.03</td>
 <td>494.20</td>
 <td>105.63</td>
 <td>13542.94</td>
@@ -869,7 +869,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>3170.84</td>
+<td>3.17</td>
 <td>481.50</td>
 <td>105.13</td>
 <td>14179.97</td>
@@ -882,7 +882,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="2">PP_StructureV3-full</td>
 <td>Intel 8350C + A100</td>
-<td>8917.95</td>
+<td>8.92</td>
 <td>697.30</td>
 <td>102.88</td>
 <td>13777.07</td>
@@ -894,7 +894,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>13115.08</td>
+<td>13.12</td>
 <td>437.40</td>
 <td>102.36</td>
 <td>13974.00</td>
@@ -907,7 +907,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="5">PP_StructureV3-seal</td>
 <td>Intel 8350C + A100</td>
-<td>1392.53</td>
+<td>1.39</td>
 <td>747.50</td>
 <td>112.55</td>
 <td>5788.79</td>
@@ -919,7 +919,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2437.59</td>
+<td>2.44</td>
 <td>630.10</td>
 <td>110.18</td>
 <td>6343.39</td>
@@ -931,7 +931,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1396.03</td>
+<td>1.40</td>
 <td>792.20</td>
 <td>113.63</td>
 <td>6673.60</td>
@@ -943,7 +943,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1745.09</td>
+<td>1.75</td>
 <td>422.40</td>
 <td>106.08</td>
 <td>6068.87</td>
@@ -955,7 +955,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>3764.94</td>
+<td>3.76</td>
 <td>400.30</td>
 <td>105.10</td>
 <td>6296.28</td>
@@ -968,7 +968,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="4">PP_StructureV3-chart</td>
 <td>Intel 8350C + A100</td>
-<td>7700.18</td>
+<td>7.70</td>
 <td>746.80</td>
 <td>102.69</td>
 <td>6355.58</td>
@@ -980,7 +980,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>10578.29</td>
+<td>10.58</td>
 <td>599.20</td>
 <td>102.51</td>
 <td>5754.14</td>
@@ -992,7 +992,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>8026.47</td>
+<td>8.03</td>
 <td>413.30</td>
 <td>101.31</td>
 <td>6473.29</td>
@@ -1004,7 +1004,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>11694.21</td>
+<td>11.69</td>
 <td>460.90</td>
 <td>101.85</td>
 <td>6503.12</td>
@@ -1017,7 +1017,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="5">PP_StructureV3-notable</td>
 <td>Intel 8350C + A100</td>
-<td>1235.49</td>
+<td>1.24</td>
 <td>738.30</td>
 <td>110.45</td>
 <td>5638.16</td>
@@ -1029,7 +1029,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2244.84</td>
+<td>2.24</td>
 <td>452.40</td>
 <td>107.79</td>
 <td>5579.15</td>
@@ -1041,7 +1041,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1182.50</td>
+<td>1.18</td>
 <td>989.00</td>
 <td>107.71</td>
 <td>6041.76</td>
@@ -1053,7 +1053,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1575.72</td>
+<td>1.58</td>
 <td>225.00</td>
 <td>102.56</td>
 <td>5518.10</td>
@@ -1065,7 +1065,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>3399.03</td>
+<td>3.40</td>
 <td>413.30</td>
 <td>103.58</td>
 <td>5874.88</td>
@@ -1078,7 +1078,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="7">PP_StructureV3-noformula</td>
 <td>Intel 6271C</td>
-<td>7846.98</td>
+<td>7.85</td>
 <td>1172.50</td>
 <td>964.70</td>
 <td>17739.00</td>
@@ -1090,7 +1090,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>8834.14</td>
+<td>8.83</td>
 <td>1053.50</td>
 <td>970.64</td>
 <td>15463.48</td>
@@ -1102,7 +1102,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>839.16</td>
+<td>0.84</td>
 <td>788.60</td>
 <td>124.25</td>
 <td>6246.39</td>
@@ -1114,7 +1114,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1417.36</td>
+<td>1.42</td>
 <td>606.20</td>
 <td>115.53</td>
 <td>7015.57</td>
@@ -1126,7 +1126,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>874.17</td>
+<td>0.87</td>
 <td>644.10</td>
 <td>119.23</td>
 <td>6895.76</td>
@@ -1138,7 +1138,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1027.28</td>
+<td>1.03</td>
 <td>377.50</td>
 <td>106.87</td>
 <td>5819.88</td>
@@ -1150,7 +1150,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>2022.90</td>
+<td>2.02</td>
 <td>430.20</td>
 <td>109.21</td>
 <td>6600.62</td>
@@ -1163,7 +1163,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">PP_StructureV3-lightweight</td>
 <td>Intel 6271C</td>
-<td>4361.78</td>
+<td>4.36</td>
 <td>1189.70</td>
 <td>995.78</td>
 <td>14000.50</td>
@@ -1175,7 +1175,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3735.85</td>
+<td>3.74</td>
 <td>1049.60</td>
 <td>967.77</td>
 <td>12960.96</td>
@@ -1187,7 +1187,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>859.69</td>
+<td>0.86</td>
 <td>572.20</td>
 <td>120.84</td>
 <td>8290.49</td>
@@ -1199,7 +1199,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>606.65</td>
+<td>0.61</td>
 <td>823.40</td>
 <td>126.25</td>
 <td>9258.22</td>
@@ -1211,7 +1211,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1067.44</td>
+<td>1.07</td>
 <td>686.80</td>
 <td>116.70</td>
 <td>9381.75</td>
@@ -1223,7 +1223,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>464.75</td>
+<td>0.46</td>
 <td>999.00</td>
 <td>122.21</td>
 <td>9734.78</td>
@@ -1235,7 +1235,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>698.54</td>
+<td>0.70</td>
 <td>355.40</td>
 <td>111.51</td>
 <td>9415.45</td>
@@ -1247,7 +1247,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>12220.94</td>
+<td>12.22</td>
 <td>223.60</td>
 <td>107.35</td>
 <td>9531.22</td>
@@ -1259,7 +1259,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1134.45</td>
+<td>1.13</td>
 <td>461.40</td>
 <td>112.16</td>
 <td>7923.09</td>
@@ -1314,7 +1314,10 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing 280 images including tables, seals, formulas, and charts.
+* Test strategy:
+    * Warm up with 20 samples, then repeat the full dataset once for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/ocr_pipelines/PP-StructureV3.md

@@ -746,11 +746,11 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="5">PP_StructureV3-default</td>
 <td>Intel 8350C + A100</td>
-<td>1378.46</td>
+<td>1.38</td>
 <td>1384.60</td>
 <td>113.26</td>
 <td>5781.59</td>
@@ -762,7 +762,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2382.03</td>
+<td>2.38</td>
 <td>608.70</td>
 <td>109.96</td>
 <td>6388.91</td>
@@ -774,7 +774,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1364.42</td>
+<td>1.36</td>
 <td>744.30</td>
 <td>112.82</td>
 <td>6199.01</td>
@@ -786,7 +786,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1735.04</td>
+<td>1.74</td>
 <td>418.50</td>
 <td>105.96</td>
 <td>6138.25</td>
@@ -798,7 +798,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>3700.82</td>
+<td>3.70</td>
 <td>434.40</td>
 <td>105.45</td>
 <td>6865.87</td>
@@ -811,7 +811,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="3">PP_StructureV3-pp</td>
 <td>Intel 8350C + A100</td>
-<td>3498.87</td>
+<td>3.50</td>
 <td>679.30</td>
 <td>105.96</td>
 <td>13850.20</td>
@@ -823,7 +823,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>5030.33</td>
+<td>5.03</td>
 <td>494.20</td>
 <td>105.63</td>
 <td>13542.94</td>
@@ -835,7 +835,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>3170.84</td>
+<td>3.17</td>
 <td>481.50</td>
 <td>105.13</td>
 <td>14179.97</td>
@@ -848,7 +848,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="2">PP_StructureV3-full</td>
 <td>Intel 8350C + A100</td>
-<td>8917.95</td>
+<td>8.92</td>
 <td>697.30</td>
 <td>102.88</td>
 <td>13777.07</td>
@@ -860,7 +860,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>13115.08</td>
+<td>13.12</td>
 <td>437.40</td>
 <td>102.36</td>
 <td>13974.00</td>
@@ -873,7 +873,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="5">PP_StructureV3-seal</td>
 <td>Intel 8350C + A100</td>
-<td>1392.53</td>
+<td>1.39</td>
 <td>747.50</td>
 <td>112.55</td>
 <td>5788.79</td>
@@ -885,7 +885,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2437.59</td>
+<td>2.44</td>
 <td>630.10</td>
 <td>110.18</td>
 <td>6343.39</td>
@@ -897,7 +897,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1396.03</td>
+<td>1.40</td>
 <td>792.20</td>
 <td>113.63</td>
 <td>6673.60</td>
@@ -909,7 +909,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1745.09</td>
+<td>1.75</td>
 <td>422.40</td>
 <td>106.08</td>
 <td>6068.87</td>
@@ -921,7 +921,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>3764.94</td>
+<td>3.76</td>
 <td>400.30</td>
 <td>105.10</td>
 <td>6296.28</td>
@@ -934,7 +934,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="4">PP_StructureV3-chart</td>
 <td>Intel 8350C + A100</td>
-<td>7700.18</td>
+<td>7.70</td>
 <td>746.80</td>
 <td>102.69</td>
 <td>6355.58</td>
@@ -946,7 +946,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>10578.29</td>
+<td>10.58</td>
 <td>599.20</td>
 <td>102.51</td>
 <td>5754.14</td>
@@ -958,7 +958,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>8026.47</td>
+<td>8.03</td>
 <td>413.30</td>
 <td>101.31</td>
 <td>6473.29</td>
@@ -970,7 +970,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>11694.21</td>
+<td>11.69</td>
 <td>460.90</td>
 <td>101.85</td>
 <td>6503.12</td>
@@ -983,7 +983,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="5">PP_StructureV3-notable</td>
 <td>Intel 8350C + A100</td>
-<td>1235.49</td>
+<td>1.24</td>
 <td>738.30</td>
 <td>110.45</td>
 <td>5638.16</td>
@@ -995,7 +995,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2244.84</td>
+<td>2.24</td>
 <td>452.40</td>
 <td>107.79</td>
 <td>5579.15</td>
@@ -1007,7 +1007,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1182.50</td>
+<td>1.18</td>
 <td>989.00</td>
 <td>107.71</td>
 <td>6041.76</td>
@@ -1019,7 +1019,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1575.72</td>
+<td>1.58</td>
 <td>225.00</td>
 <td>102.56</td>
 <td>5518.10</td>
@@ -1031,7 +1031,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>3399.03</td>
+<td>3.40</td>
 <td>413.30</td>
 <td>103.58</td>
 <td>5874.88</td>
@@ -1044,7 +1044,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="7">PP_StructureV3-noformula</td>
 <td>Intel 6271C</td>
-<td>7846.98</td>
+<td>7.85</td>
 <td>1172.50</td>
 <td>964.70</td>
 <td>17739.00</td>
@@ -1056,7 +1056,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>8834.14</td>
+<td>8.83</td>
 <td>1053.50</td>
 <td>970.64</td>
 <td>15463.48</td>
@@ -1068,7 +1068,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>839.16</td>
+<td>0.84</td>
 <td>788.60</td>
 <td>124.25</td>
 <td>6246.39</td>
@@ -1080,7 +1080,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1417.36</td>
+<td>1.42</td>
 <td>606.20</td>
 <td>115.53</td>
 <td>7015.57</td>
@@ -1092,7 +1092,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>874.17</td>
+<td>0.87</td>
 <td>644.10</td>
 <td>119.23</td>
 <td>6895.76</td>
@@ -1104,7 +1104,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1027.28</td>
+<td>1.03</td>
 <td>377.50</td>
 <td>106.87</td>
 <td>5819.88</td>
@@ -1116,7 +1116,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>2022.90</td>
+<td>2.02</td>
 <td>430.20</td>
 <td>109.21</td>
 <td>6600.62</td>
@@ -1129,7 +1129,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">PP_StructureV3-lightweight</td>
 <td>Intel 6271C</td>
-<td>4361.78</td>
+<td>4.36</td>
 <td>1189.70</td>
 <td>995.78</td>
 <td>14000.50</td>
@@ -1141,7 +1141,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3735.85</td>
+<td>3.74</td>
 <td>1049.60</td>
 <td>967.77</td>
 <td>12960.96</td>
@@ -1153,7 +1153,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>859.69</td>
+<td>0.86</td>
 <td>572.20</td>
 <td>120.84</td>
 <td>8290.49</td>
@@ -1165,7 +1165,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>606.65</td>
+<td>0.61</td>
 <td>823.40</td>
 <td>126.25</td>
 <td>9258.22</td>
@@ -1177,7 +1177,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1067.44</td>
+<td>1.07</td>
 <td>686.80</td>
 <td>116.70</td>
 <td>9381.75</td>
@@ -1189,7 +1189,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>464.75</td>
+<td>0.46</td>
 <td>999.00</td>
 <td>122.21</td>
 <td>9734.78</td>
@@ -1201,7 +1201,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>698.54</td>
+<td>0.70</td>
 <td>355.40</td>
 <td>111.51</td>
 <td>9415.45</td>
@@ -1213,7 +1213,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>12220.94</td>
+<td>12.22</td>
 <td>223.60</td>
 <td>107.35</td>
 <td>9531.22</td>
@@ -1225,7 +1225,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1134.45</td>
+<td>1.13</td>
 <td>461.40</td>
 <td>112.16</td>
 <td>7923.09</td>
@@ -1280,7 +1280,10 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含表格、印章、公式、图表的280张图像。
+* 测试策略：
+    * 使用 20 个样本进行预热，然后对整个数据集重复 1 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
 

docs/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.en.md

@@ -127,11 +127,11 @@ The document image preprocessing pipeline integrates two major functions: docume
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
 <td rowspan="9">doc_preprocessor-default</td>
 <td>Intel 6271C</td>
-<td>1127.85</td>
+<td>1.13</td>
 <td>1012.50</td>
 <td>789.07</td>
 <td>1913.89</td>
@@ -143,7 +143,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>764.23</td>
+<td>0.76</td>
 <td>1003.70</td>
 <td>782.77</td>
 <td>1916.81</td>
@@ -155,7 +155,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>96.74</td>
+<td>0.10</td>
 <td>174.70</td>
 <td>135.91</td>
 <td>2023.54</td>
@@ -167,7 +167,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>76.66</td>
+<td>0.08</td>
 <td>148.90</td>
 <td>125.73</td>
 <td>1748.06</td>
@@ -179,7 +179,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>114.96</td>
+<td>0.11</td>
 <td>168.60</td>
 <td>131.05</td>
 <td>1837.66</td>
@@ -191,7 +191,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>69.34</td>
+<td>0.07</td>
 <td>140.90</td>
 <td>122.80</td>
 <td>1992.86</td>
@@ -203,7 +203,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>81.65</td>
+<td>0.08</td>
 <td>146.40</td>
 <td>125.37</td>
 <td>1896.46</td>
@@ -215,7 +215,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>M4</td>
-<td>368.93</td>
+<td>0.37</td>
 <td>118.20</td>
 <td>103.40</td>
 <td>1973.23</td>
@@ -227,7 +227,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>122.40</td>
+<td>0.12</td>
 <td>156.90</td>
 <td>123.75</td>
 <td>1874.29</td>
@@ -240,7 +240,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 <tr>
 <td rowspan="9">doc_preprocessor-clsonly</td>
 <td>Intel 6271C</td>
-<td>55.12</td>
+<td>0.06</td>
 <td>1085.60</td>
 <td>1057.12</td>
 <td>1254.27</td>
@@ -252,7 +252,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>59.48</td>
+<td>0.06</td>
 <td>1051.60</td>
 <td>1043.39</td>
 <td>1263.88</td>
@@ -264,7 +264,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>47.44</td>
+<td>0.05</td>
 <td>202.70</td>
 <td>173.75</td>
 <td>1868.11</td>
@@ -276,7 +276,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>43.27</td>
+<td>0.04</td>
 <td>167.80</td>
 <td>150.56</td>
 <td>1347.92</td>
@@ -288,7 +288,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>53.72</td>
+<td>0.05</td>
 <td>202.70</td>
 <td>170.72</td>
 <td>1404.16</td>
@@ -300,7 +300,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>41.89</td>
+<td>0.04</td>
 <td>153.80</td>
 <td>141.46</td>
 <td>1590.48</td>
@@ -312,7 +312,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>44.61</td>
+<td>0.04</td>
 <td>179.80</td>
 <td>156.96</td>
 <td>1569.85</td>
@@ -324,7 +324,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>M4</td>
-<td>34.12</td>
+<td>0.03</td>
 <td>127.60</td>
 <td>123.24</td>
 <td>1459.06</td>
@@ -336,7 +336,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>51.81</td>
+<td>0.05</td>
 <td>180.80</td>
 <td>154.34</td>
 <td>1515.43</td>
@@ -349,7 +349,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 <tr>
 <td rowspan="9">doc_preprocessor-unwarponly</td>
 <td>Intel 6271C</td>
-<td>1092.82</td>
+<td>1.09</td>
 <td>1002.10</td>
 <td>749.82</td>
 <td>1875.45</td>
@@ -361,7 +361,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>747.32</td>
+<td>0.75</td>
 <td>951.30</td>
 <td>754.01</td>
 <td>1843.54</td>
@@ -373,7 +373,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>95.22</td>
+<td>0.10</td>
 <td>101.20</td>
 <td>100.08</td>
 <td>1960.48</td>
@@ -385,7 +385,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>78.05</td>
+<td>0.08</td>
 <td>100.90</td>
 <td>100.15</td>
 <td>1570.25</td>
@@ -397,7 +397,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>130.84</td>
+<td>0.13</td>
 <td>100.90</td>
 <td>100.25</td>
 <td>1561.52</td>
@@ -409,7 +409,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>68.62</td>
+<td>0.07</td>
 <td>106.90</td>
 <td>100.62</td>
 <td>1808.00</td>
@@ -421,7 +421,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>84.19</td>
+<td>0.08</td>
 <td>101.90</td>
 <td>100.14</td>
 <td>1848.25</td>
@@ -433,7 +433,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>M4</td>
-<td>295.15</td>
+<td>0.30</td>
 <td>102.40</td>
 <td>100.76</td>
 <td>1957.42</td>
@@ -445,7 +445,7 @@ The document image preprocessing pipeline integrates two major functions: docume
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>117.25</td>
+<td>0.12</td>
 <td>101.80</td>
 <td>100.36</td>
 <td>1674.25</td>
@@ -480,7 +480,10 @@ The document image preprocessing pipeline integrates two major functions: docume
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing 34 images for document orientation classification and image distortion.
+* Test strategy:
+    * Warm up with 20 samples, then repeat the full dataset 5 times for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.md

@@ -126,11 +126,11 @@ comments: true
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="9">doc_preprocessor-default</td>
 <td>Intel 6271C</td>
-<td>1127.85</td>
+<td>1.13</td>
 <td>1012.50</td>
 <td>789.07</td>
 <td>1913.89</td>
@@ -142,7 +142,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>764.23</td>
+<td>0.76</td>
 <td>1003.70</td>
 <td>782.77</td>
 <td>1916.81</td>
@@ -154,7 +154,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>96.74</td>
+<td>0.10</td>
 <td>174.70</td>
 <td>135.91</td>
 <td>2023.54</td>
@@ -166,7 +166,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>76.66</td>
+<td>0.08</td>
 <td>148.90</td>
 <td>125.73</td>
 <td>1748.06</td>
@@ -178,7 +178,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>114.96</td>
+<td>0.11</td>
 <td>168.60</td>
 <td>131.05</td>
 <td>1837.66</td>
@@ -190,7 +190,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>69.34</td>
+<td>0.07</td>
 <td>140.90</td>
 <td>122.80</td>
 <td>1992.86</td>
@@ -202,7 +202,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>81.65</td>
+<td>0.08</td>
 <td>146.40</td>
 <td>125.37</td>
 <td>1896.46</td>
@@ -214,7 +214,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>368.93</td>
+<td>0.37</td>
 <td>118.20</td>
 <td>103.40</td>
 <td>1973.23</td>
@@ -226,7 +226,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>122.40</td>
+<td>0.12</td>
 <td>156.90</td>
 <td>123.75</td>
 <td>1874.29</td>
@@ -239,7 +239,7 @@ comments: true
 <tr>
 <td rowspan="9">doc_preprocessor-clsonly</td>
 <td>Intel 6271C</td>
-<td>55.12</td>
+<td>0.06</td>
 <td>1085.60</td>
 <td>1057.12</td>
 <td>1254.27</td>
@@ -251,7 +251,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>59.48</td>
+<td>0.06</td>
 <td>1051.60</td>
 <td>1043.39</td>
 <td>1263.88</td>
@@ -263,7 +263,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>47.44</td>
+<td>0.05</td>
 <td>202.70</td>
 <td>173.75</td>
 <td>1868.11</td>
@@ -275,7 +275,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>43.27</td>
+<td>0.04</td>
 <td>167.80</td>
 <td>150.56</td>
 <td>1347.92</td>
@@ -287,7 +287,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>53.72</td>
+<td>0.05</td>
 <td>202.70</td>
 <td>170.72</td>
 <td>1404.16</td>
@@ -299,7 +299,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>41.89</td>
+<td>0.04</td>
 <td>153.80</td>
 <td>141.46</td>
 <td>1590.48</td>
@@ -311,7 +311,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>44.61</td>
+<td>0.04</td>
 <td>179.80</td>
 <td>156.96</td>
 <td>1569.85</td>
@@ -323,7 +323,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>34.12</td>
+<td>0.03</td>
 <td>127.60</td>
 <td>123.24</td>
 <td>1459.06</td>
@@ -335,7 +335,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>51.81</td>
+<td>0.05</td>
 <td>180.80</td>
 <td>154.34</td>
 <td>1515.43</td>
@@ -348,7 +348,7 @@ comments: true
 <tr>
 <td rowspan="9">doc_preprocessor-unwarponly</td>
 <td>Intel 6271C</td>
-<td>1092.82</td>
+<td>1.09</td>
 <td>1002.10</td>
 <td>749.82</td>
 <td>1875.45</td>
@@ -360,7 +360,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>747.32</td>
+<td>0.75</td>
 <td>951.30</td>
 <td>754.01</td>
 <td>1843.54</td>
@@ -372,7 +372,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>95.22</td>
+<td>0.10</td>
 <td>101.20</td>
 <td>100.08</td>
 <td>1960.48</td>
@@ -384,7 +384,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>78.05</td>
+<td>0.08</td>
 <td>100.90</td>
 <td>100.15</td>
 <td>1570.25</td>
@@ -396,7 +396,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>130.84</td>
+<td>0.13</td>
 <td>100.90</td>
 <td>100.25</td>
 <td>1561.52</td>
@@ -408,7 +408,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>68.62</td>
+<td>0.07</td>
 <td>106.90</td>
 <td>100.62</td>
 <td>1808.00</td>
@@ -420,7 +420,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>84.19</td>
+<td>0.08</td>
 <td>101.90</td>
 <td>100.14</td>
 <td>1848.25</td>
@@ -432,7 +432,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>295.15</td>
+<td>0.30</td>
 <td>102.40</td>
 <td>100.76</td>
 <td>1957.42</td>
@@ -444,7 +444,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>117.25</td>
+<td>0.12</td>
 <td>101.80</td>
 <td>100.36</td>
 <td>1674.25</td>
@@ -479,7 +479,10 @@ comments: true
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含文档方向分类和图像形变的34张图像。
+* 测试策略：
+    * 使用 20 个样本进行预热，然后对整个数据集重复 5 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
 

docs/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.en.md

@@ -406,11 +406,11 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
 <td rowspan="9">formula_recognition-default</td>
 <td>Intel 6271C</td>
-<td>26270.51</td>
+<td>26.27</td>
 <td>1039.70</td>
 <td>1003.86</td>
 <td>3093.11</td>
@@ -422,7 +422,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>18242.94</td>
+<td>18.24</td>
 <td>1048.60</td>
 <td>893.64</td>
 <td>3029.64</td>
@@ -434,7 +434,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>5971.78</td>
+<td>5.97</td>
 <td>120.80</td>
 <td>101.63</td>
 <td>2255.59</td>
@@ -446,7 +446,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>1441.69</td>
+<td>1.44</td>
 <td>111.60</td>
 <td>103.08</td>
 <td>2145.61</td>
@@ -458,7 +458,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2028.69</td>
+<td>2.03</td>
 <td>129.80</td>
 <td>104.93</td>
 <td>2231.39</td>
@@ -470,7 +470,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1167.13</td>
+<td>1.17</td>
 <td>109.90</td>
 <td>102.93</td>
 <td>2509.55</td>
@@ -482,7 +482,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1581.55</td>
+<td>1.58</td>
 <td>115.90</td>
 <td>103.61</td>
 <td>2442.22</td>
@@ -494,7 +494,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>10159.43</td>
+<td>10.16</td>
 <td>126.60</td>
 <td>107.78</td>
 <td>3135.38</td>
@@ -506,7 +506,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>2420.85</td>
+<td>2.42</td>
 <td>115.00</td>
 <td>102.81</td>
 <td>2350.60</td>
@@ -519,7 +519,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <tr>
 <td rowspan="9">formula_recognition-nopp</td>
 <td>Intel 6271C</td>
-<td>24857.99</td>
+<td>24.86</td>
 <td>1050.30</td>
 <td>1009.22</td>
 <td>2715.87</td>
@@ -531,7 +531,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>17002.42</td>
+<td>17.00</td>
 <td>1046.60</td>
 <td>891.66</td>
 <td>2701.11</td>
@@ -543,7 +543,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>2428.21</td>
+<td>2.43</td>
 <td>112.90</td>
 <td>102.75</td>
 <td>1972.57</td>
@@ -555,7 +555,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>1363.52</td>
+<td>1.36</td>
 <td>106.90</td>
 <td>102.16</td>
 <td>2167.37</td>
@@ -567,7 +567,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1933.41</td>
+<td>1.93</td>
 <td>112.30</td>
 <td>103.38</td>
 <td>1878.44</td>
@@ -579,7 +579,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1089.64</td>
+<td>1.09</td>
 <td>108.80</td>
 <td>102.04</td>
 <td>2276.06</td>
@@ -591,7 +591,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1457.44</td>
+<td>1.46</td>
 <td>106.30</td>
 <td>102.23</td>
 <td>1972.24</td>
@@ -603,7 +603,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>13516.34</td>
+<td>13.52</td>
 <td>131.60</td>
 <td>107.52</td>
 <td>2502.56</td>
@@ -615,7 +615,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>2273.27</td>
+<td>2.27</td>
 <td>108.00</td>
 <td>101.84</td>
 <td>1942.30</td>
@@ -628,7 +628,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <tr>
 <td rowspan="8">formula_recognition-nolayout</td>
 <td>Intel 8350C</td>
-<td>65371.52</td>
+<td>65.37</td>
 <td>1068.60</td>
 <td>1022.44</td>
 <td>2566.70</td>
@@ -640,7 +640,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>16886.29</td>
+<td>16.89</td>
 <td>118.90</td>
 <td>100.37</td>
 <td>2239.15</td>
@@ -652,7 +652,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>10305.23</td>
+<td>10.31</td>
 <td>109.90</td>
 <td>100.34</td>
 <td>2139.89</td>
@@ -664,7 +664,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>15867.71</td>
+<td>15.87</td>
 <td>118.80</td>
 <td>100.76</td>
 <td>2280.21</td>
@@ -676,7 +676,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>6451.56</td>
+<td>6.45</td>
 <td>109.90</td>
 <td>100.38</td>
 <td>2348.35</td>
@@ -688,7 +688,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>10484.44</td>
+<td>10.48</td>
 <td>113.90</td>
 <td>100.44</td>
 <td>2348.49</td>
@@ -700,7 +700,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>42118.26</td>
+<td>42.12</td>
 <td>149.50</td>
 <td>128.97</td>
 <td>2437.94</td>
@@ -712,7 +712,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>13954.69</td>
+<td>13.95</td>
 <td>114.90</td>
 <td>100.56</td>
 <td>2346.36</td>
@@ -725,7 +725,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <tr>
 <td rowspan="8">formula_recognition-nopp-nolayout</td>
 <td>Intel 8350C</td>
-<td>56834.33</td>
+<td>56.83</td>
 <td>1067.80</td>
 <td>1023.31</td>
 <td>2173.72</td>
@@ -737,7 +737,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>16577.80</td>
+<td>16.58</td>
 <td>104.90</td>
 <td>100.18</td>
 <td>2017.30</td>
@@ -749,7 +749,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>9084.45</td>
+<td>9.08</td>
 <td>102.90</td>
 <td>100.19</td>
 <td>1747.94</td>
@@ -761,7 +761,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>13967.21</td>
+<td>13.97</td>
 <td>116.80</td>
 <td>100.45</td>
 <td>1864.97</td>
@@ -773,7 +773,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>6454.01</td>
+<td>6.45</td>
 <td>107.00</td>
 <td>100.19</td>
 <td>2163.71</td>
@@ -785,7 +785,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>9151.06</td>
+<td>9.15</td>
 <td>103.00</td>
 <td>100.24</td>
 <td>1998.28</td>
@@ -797,7 +797,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>39774.40</td>
+<td>39.77</td>
 <td>149.80</td>
 <td>129.49</td>
 <td>2098.33</td>
@@ -809,7 +809,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>11566.69</td>
+<td>11.57</td>
 <td>103.00</td>
 <td>100.37</td>
 <td>1995.27</td>
@@ -822,7 +822,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <tr>
 <td rowspan="9">formula_recognition-nopp-lightweight</td>
 <td>Intel 6271C</td>
-<td>6538.36</td>
+<td>6.54</td>
 <td>1023.60</td>
 <td>1007.63</td>
 <td>2097.84</td>
@@ -834,7 +834,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>4720.90</td>
+<td>4.72</td>
 <td>1028.60</td>
 <td>926.98</td>
 <td>2089.03</td>
@@ -846,7 +846,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>761.03</td>
+<td>0.76</td>
 <td>116.90</td>
 <td>107.99</td>
 <td>1909.78</td>
@@ -858,7 +858,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>489.65</td>
+<td>0.49</td>
 <td>109.90</td>
 <td>105.51</td>
 <td>1789.37</td>
@@ -870,7 +870,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>665.85</td>
+<td>0.67</td>
 <td>116.80</td>
 <td>109.35</td>
 <td>1751.75</td>
@@ -882,7 +882,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>417.60</td>
+<td>0.42</td>
 <td>112.90</td>
 <td>105.16</td>
 <td>2249.38</td>
@@ -894,7 +894,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>521.10</td>
+<td>0.52</td>
 <td>108.90</td>
 <td>105.69</td>
 <td>1903.01</td>
@@ -906,7 +906,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>6153.89</td>
+<td>6.15</td>
 <td>116.00</td>
 <td>105.42</td>
 <td>2048.00</td>
@@ -918,7 +918,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>720.11</td>
+<td>0.72</td>
 <td>110.50</td>
 <td>105.53</td>
 <td>1912.62</td>
@@ -931,7 +931,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <tr>
 <td rowspan="9">formula_recognition-nopp-lightweightlayout</td>
 <td>Intel 6271C</td>
-<td>7562.69</td>
+<td>7.56</td>
 <td>1052.60</td>
 <td>1015.57</td>
 <td>2214.38</td>
@@ -943,7 +943,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>5462.90</td>
+<td>5.46</td>
 <td>1056.60</td>
 <td>940.86</td>
 <td>2217.84</td>
@@ -955,7 +955,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>1264.31</td>
+<td>1.26</td>
 <td>115.80</td>
 <td>103.72</td>
 <td>1937.91</td>
@@ -967,7 +967,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>695.32</td>
+<td>0.70</td>
 <td>107.90</td>
 <td>102.54</td>
 <td>1740.03</td>
@@ -979,7 +979,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1014.42</td>
+<td>1.01</td>
 <td>114.00</td>
 <td>104.15</td>
 <td>1760.52</td>
@@ -991,7 +991,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>561.21</td>
+<td>0.56</td>
 <td>107.80</td>
 <td>102.41</td>
 <td>2073.04</td>
@@ -1003,7 +1003,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>727.13</td>
+<td>0.73</td>
 <td>108.90</td>
 <td>102.77</td>
 <td>1952.25</td>
@@ -1015,7 +1015,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>2866.01</td>
+<td>2.87</td>
 <td>124.10</td>
 <td>113.46</td>
 <td>2467.97</td>
@@ -1027,7 +1027,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1042.95</td>
+<td>1.04</td>
 <td>108.00</td>
 <td>102.42</td>
 <td>1928.40</td>
@@ -1040,7 +1040,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 <tr>
 <td rowspan="9">formula_recognition-nopp-nolayout-lightweight</td>
 <td>Intel 6271C</td>
-<td>2798.43</td>
+<td>2.80</td>
 <td>1047.60</td>
 <td>1031.25</td>
 <td>1534.17</td>
@@ -1052,7 +1052,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>2315.28</td>
+<td>2.32</td>
 <td>1052.60</td>
 <td>1034.39</td>
 <td>1523.69</td>
@@ -1064,7 +1064,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>1000.18</td>
+<td>1.00</td>
 <td>104.90</td>
 <td>101.83</td>
 <td>1859.66</td>
@@ -1076,7 +1076,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>586.57</td>
+<td>0.59</td>
 <td>103.90</td>
 <td>102.00</td>
 <td>1633.84</td>
@@ -1088,7 +1088,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>792.55</td>
+<td>0.79</td>
 <td>108.90</td>
 <td>103.25</td>
 <td>1658.45</td>
@@ -1100,7 +1100,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>473.62</td>
+<td>0.47</td>
 <td>108.00</td>
 <td>102.07</td>
 <td>2023.42</td>
@@ -1112,7 +1112,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>623.90</td>
+<td>0.62</td>
 <td>103.00</td>
 <td>101.00</td>
 <td>1918.49</td>
@@ -1124,7 +1124,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>M4</td>
-<td>1567.94</td>
+<td>1.57</td>
 <td>119.50</td>
 <td>117.72</td>
 <td>1489.78</td>
@@ -1136,7 +1136,7 @@ The formula recognition pipeline is designed to solve formula recognition tasks
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>825.00</td>
+<td>0.83</td>
 <td>105.00</td>
 <td>101.98</td>
 <td>1934.31</td>
@@ -1187,7 +1187,10 @@ The formula recognition pipeline is designed to solve formula recognition tasks
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing 4 images and 1 PDF, including Chinese document formulas, English document formulas, standalone Chinese formulas, and tables with embedded formulas.
+* Test strategy:
+    * Warm up with 3 samples, then repeat the full dataset 20 times for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.md

@@ -408,11 +408,11 @@ comments: true
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="9">formula_recognition-default</td>
 <td>Intel 6271C</td>
-<td>26270.51</td>
+<td>26.27</td>
 <td>1039.70</td>
 <td>1003.86</td>
 <td>3093.11</td>
@@ -424,7 +424,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>18242.94</td>
+<td>18.24</td>
 <td>1048.60</td>
 <td>893.64</td>
 <td>3029.64</td>
@@ -436,7 +436,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>5971.78</td>
+<td>5.97</td>
 <td>120.80</td>
 <td>101.63</td>
 <td>2255.59</td>
@@ -448,7 +448,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>1441.69</td>
+<td>1.44</td>
 <td>111.60</td>
 <td>103.08</td>
 <td>2145.61</td>
@@ -460,7 +460,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>2028.69</td>
+<td>2.03</td>
 <td>129.80</td>
 <td>104.93</td>
 <td>2231.39</td>
@@ -472,7 +472,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1167.13</td>
+<td>1.17</td>
 <td>109.90</td>
 <td>102.93</td>
 <td>2509.55</td>
@@ -484,7 +484,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1581.55</td>
+<td>1.58</td>
 <td>115.90</td>
 <td>103.61</td>
 <td>2442.22</td>
@@ -496,7 +496,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>10159.43</td>
+<td>10.16</td>
 <td>126.60</td>
 <td>107.78</td>
 <td>3135.38</td>
@@ -508,7 +508,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>2420.85</td>
+<td>2.42</td>
 <td>115.00</td>
 <td>102.81</td>
 <td>2350.60</td>
@@ -521,7 +521,7 @@ comments: true
 <tr>
 <td rowspan="9">formula_recognition-nopp</td>
 <td>Intel 6271C</td>
-<td>24857.99</td>
+<td>24.86</td>
 <td>1050.30</td>
 <td>1009.22</td>
 <td>2715.87</td>
@@ -533,7 +533,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>17002.42</td>
+<td>17.00</td>
 <td>1046.60</td>
 <td>891.66</td>
 <td>2701.11</td>
@@ -545,7 +545,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>2428.21</td>
+<td>2.43</td>
 <td>112.90</td>
 <td>102.75</td>
 <td>1972.57</td>
@@ -557,7 +557,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>1363.52</td>
+<td>1.36</td>
 <td>106.90</td>
 <td>102.16</td>
 <td>2167.37</td>
@@ -569,7 +569,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1933.41</td>
+<td>1.93</td>
 <td>112.30</td>
 <td>103.38</td>
 <td>1878.44</td>
@@ -581,7 +581,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>1089.64</td>
+<td>1.09</td>
 <td>108.80</td>
 <td>102.04</td>
 <td>2276.06</td>
@@ -593,7 +593,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>1457.44</td>
+<td>1.46</td>
 <td>106.30</td>
 <td>102.23</td>
 <td>1972.24</td>
@@ -605,7 +605,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>13516.34</td>
+<td>13.52</td>
 <td>131.60</td>
 <td>107.52</td>
 <td>2502.56</td>
@@ -617,7 +617,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>2273.27</td>
+<td>2.27</td>
 <td>108.00</td>
 <td>101.84</td>
 <td>1942.30</td>
@@ -630,7 +630,7 @@ comments: true
 <tr>
 <td rowspan="8">formula_recognition-nolayout</td>
 <td>Intel 8350C</td>
-<td>65371.52</td>
+<td>65.37</td>
 <td>1068.60</td>
 <td>1022.44</td>
 <td>2566.70</td>
@@ -642,7 +642,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>16886.29</td>
+<td>16.89</td>
 <td>118.90</td>
 <td>100.37</td>
 <td>2239.15</td>
@@ -654,7 +654,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>10305.23</td>
+<td>10.31</td>
 <td>109.90</td>
 <td>100.34</td>
 <td>2139.89</td>
@@ -666,7 +666,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>15867.71</td>
+<td>15.87</td>
 <td>118.80</td>
 <td>100.76</td>
 <td>2280.21</td>
@@ -678,7 +678,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>6451.56</td>
+<td>6.45</td>
 <td>109.90</td>
 <td>100.38</td>
 <td>2348.35</td>
@@ -690,7 +690,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>10484.44</td>
+<td>10.48</td>
 <td>113.90</td>
 <td>100.44</td>
 <td>2348.49</td>
@@ -702,7 +702,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>42118.26</td>
+<td>42.12</td>
 <td>149.50</td>
 <td>128.97</td>
 <td>2437.94</td>
@@ -714,7 +714,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>13954.69</td>
+<td>13.95</td>
 <td>114.90</td>
 <td>100.56</td>
 <td>2346.36</td>
@@ -727,7 +727,7 @@ comments: true
 <tr>
 <td rowspan="8">formula_recognition-nopp-nolayout</td>
 <td>Intel 8350C</td>
-<td>56834.33</td>
+<td>56.83</td>
 <td>1067.80</td>
 <td>1023.31</td>
 <td>2173.72</td>
@@ -739,7 +739,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>16577.80</td>
+<td>16.58</td>
 <td>104.90</td>
 <td>100.18</td>
 <td>2017.30</td>
@@ -751,7 +751,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>9084.45</td>
+<td>9.08</td>
 <td>102.90</td>
 <td>100.19</td>
 <td>1747.94</td>
@@ -763,7 +763,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>13967.21</td>
+<td>13.97</td>
 <td>116.80</td>
 <td>100.45</td>
 <td>1864.97</td>
@@ -775,7 +775,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>6454.01</td>
+<td>6.45</td>
 <td>107.00</td>
 <td>100.19</td>
 <td>2163.71</td>
@@ -787,7 +787,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>9151.06</td>
+<td>9.15</td>
 <td>103.00</td>
 <td>100.24</td>
 <td>1998.28</td>
@@ -799,7 +799,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>39774.40</td>
+<td>39.77</td>
 <td>149.80</td>
 <td>129.49</td>
 <td>2098.33</td>
@@ -811,7 +811,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>11566.69</td>
+<td>11.57</td>
 <td>103.00</td>
 <td>100.37</td>
 <td>1995.27</td>
@@ -824,7 +824,7 @@ comments: true
 <tr>
 <td rowspan="9">formula_recognition-nopp-lightweight</td>
 <td>Intel 6271C</td>
-<td>6538.36</td>
+<td>6.54</td>
 <td>1023.60</td>
 <td>1007.63</td>
 <td>2097.84</td>
@@ -836,7 +836,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>4720.90</td>
+<td>4.72</td>
 <td>1028.60</td>
 <td>926.98</td>
 <td>2089.03</td>
@@ -848,7 +848,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>761.03</td>
+<td>0.76</td>
 <td>116.90</td>
 <td>107.99</td>
 <td>1909.78</td>
@@ -860,7 +860,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>489.65</td>
+<td>0.49</td>
 <td>109.90</td>
 <td>105.51</td>
 <td>1789.37</td>
@@ -872,7 +872,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>665.85</td>
+<td>0.67</td>
 <td>116.80</td>
 <td>109.35</td>
 <td>1751.75</td>
@@ -884,7 +884,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>417.60</td>
+<td>0.42</td>
 <td>112.90</td>
 <td>105.16</td>
 <td>2249.38</td>
@@ -896,7 +896,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>521.10</td>
+<td>0.52</td>
 <td>108.90</td>
 <td>105.69</td>
 <td>1903.01</td>
@@ -908,7 +908,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>6153.89</td>
+<td>6.15</td>
 <td>116.00</td>
 <td>105.42</td>
 <td>2048.00</td>
@@ -920,7 +920,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>720.11</td>
+<td>0.72</td>
 <td>110.50</td>
 <td>105.53</td>
 <td>1912.62</td>
@@ -933,7 +933,7 @@ comments: true
 <tr>
 <td rowspan="9">formula_recognition-nopp-lightweightlayout</td>
 <td>Intel 6271C</td>
-<td>7562.69</td>
+<td>7.56</td>
 <td>1052.60</td>
 <td>1015.57</td>
 <td>2214.38</td>
@@ -945,7 +945,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>5462.90</td>
+<td>5.46</td>
 <td>1056.60</td>
 <td>940.86</td>
 <td>2217.84</td>
@@ -957,7 +957,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>1264.31</td>
+<td>1.26</td>
 <td>115.80</td>
 <td>103.72</td>
 <td>1937.91</td>
@@ -969,7 +969,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>695.32</td>
+<td>0.70</td>
 <td>107.90</td>
 <td>102.54</td>
 <td>1740.03</td>
@@ -981,7 +981,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>1014.42</td>
+<td>1.01</td>
 <td>114.00</td>
 <td>104.15</td>
 <td>1760.52</td>
@@ -993,7 +993,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>561.21</td>
+<td>0.56</td>
 <td>107.80</td>
 <td>102.41</td>
 <td>2073.04</td>
@@ -1005,7 +1005,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>727.13</td>
+<td>0.73</td>
 <td>108.90</td>
 <td>102.77</td>
 <td>1952.25</td>
@@ -1017,7 +1017,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>2866.01</td>
+<td>2.87</td>
 <td>124.10</td>
 <td>113.46</td>
 <td>2467.97</td>
@@ -1029,7 +1029,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>1042.95</td>
+<td>1.04</td>
 <td>108.00</td>
 <td>102.42</td>
 <td>1928.40</td>
@@ -1042,7 +1042,7 @@ comments: true
 <tr>
 <td rowspan="9">formula_recognition-nopp-nolayout-lightweight</td>
 <td>Intel 6271C</td>
-<td>2798.43</td>
+<td>2.80</td>
 <td>1047.60</td>
 <td>1031.25</td>
 <td>1534.17</td>
@@ -1054,7 +1054,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>2315.28</td>
+<td>2.32</td>
 <td>1052.60</td>
 <td>1034.39</td>
 <td>1523.69</td>
@@ -1066,7 +1066,7 @@ comments: true
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>1000.18</td>
+<td>1.00</td>
 <td>104.90</td>
 <td>101.83</td>
 <td>1859.66</td>
@@ -1078,7 +1078,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>586.57</td>
+<td>0.59</td>
 <td>103.90</td>
 <td>102.00</td>
 <td>1633.84</td>
@@ -1090,7 +1090,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>792.55</td>
+<td>0.79</td>
 <td>108.90</td>
 <td>103.25</td>
 <td>1658.45</td>
@@ -1102,7 +1102,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>473.62</td>
+<td>0.47</td>
 <td>108.00</td>
 <td>102.07</td>
 <td>2023.42</td>
@@ -1114,7 +1114,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>623.90</td>
+<td>0.62</td>
 <td>103.00</td>
 <td>101.00</td>
 <td>1918.49</td>
@@ -1126,7 +1126,7 @@ comments: true
 </tr>
 <tr>
 <td>M4</td>
-<td>1567.94</td>
+<td>1.57</td>
 <td>119.50</td>
 <td>117.72</td>
 <td>1489.78</td>
@@ -1138,7 +1138,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>825.00</td>
+<td>0.83</td>
 <td>105.00</td>
 <td>101.98</td>
 <td>1934.31</td>
@@ -1189,7 +1189,10 @@ comments: true
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含中文文档公式、英文文档公式、中文公式和带有公式的表格的4张图像和1个pdf。
+* 测试策略：
+    * 使用 3 个样本进行预热，然后对整个数据集重复 20 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
 

docs/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.en.md

@@ -643,11 +643,11 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
 <td rowspan="9">seal_recognition-default</td>
 <td>Intel 6271C</td>
-<td>1441.99</td>
+<td>1.44</td>
 <td>1013.50</td>
 <td>1005.77</td>
 <td>2850.27</td>
@@ -659,7 +659,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1303.84</td>
+<td>1.30</td>
 <td>1010.40</td>
 <td>970.29</td>
 <td>2797.28</td>
@@ -671,7 +671,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>102.05</td>
+<td>0.10</td>
 <td>115.80</td>
 <td>112.72</td>
 <td>1647.39</td>
@@ -683,7 +683,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>67.80</td>
+<td>0.07</td>
 <td>117.90</td>
 <td>115.03</td>
 <td>1519.09</td>
@@ -695,7 +695,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>115.01</td>
+<td>0.12</td>
 <td>117.80</td>
 <td>115.16</td>
 <td>1425.80</td>
@@ -707,7 +707,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>77.40</td>
+<td>0.08</td>
 <td>117.40</td>
 <td>111.64</td>
 <td>1721.26</td>
@@ -719,7 +719,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>79.64</td>
+<td>0.08</td>
 <td>125.80</td>
 <td>118.96</td>
 <td>1727.16</td>
@@ -731,7 +731,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>1543.47</td>
+<td>1.54</td>
 <td>104.70</td>
 <td>101.57</td>
 <td>4526.11</td>
@@ -743,7 +743,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>155.57</td>
+<td>0.16</td>
 <td>110.20</td>
 <td>108.01</td>
 <td>1690.79</td>
@@ -756,7 +756,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">seal_recognition-nopp</td>
 <td>Intel 6271C</td>
-<td>1109.72</td>
+<td>1.11</td>
 <td>1014.70</td>
 <td>1009.28</td>
 <td>3062.18</td>
@@ -768,7 +768,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1203.15</td>
+<td>1.20</td>
 <td>1012.60</td>
 <td>1006.68</td>
 <td>3079.39</td>
@@ -780,7 +780,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>119.58</td>
+<td>0.12</td>
 <td>116.90</td>
 <td>115.51</td>
 <td>1608.80</td>
@@ -792,7 +792,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>76.98</td>
+<td>0.08</td>
 <td>118.90</td>
 <td>117.49</td>
 <td>1440.84</td>
@@ -804,7 +804,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>124.81</td>
+<td>0.12</td>
 <td>119.80</td>
 <td>117.69</td>
 <td>1451.77</td>
@@ -816,7 +816,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>102.76</td>
+<td>0.10</td>
 <td>117.90</td>
 <td>111.53</td>
 <td>1722.02</td>
@@ -828,7 +828,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>95.30</td>
+<td>0.10</td>
 <td>120.80</td>
 <td>118.38</td>
 <td>1642.00</td>
@@ -840,7 +840,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>2343.35</td>
+<td>2.34</td>
 <td>104.30</td>
 <td>101.47</td>
 <td>4436.23</td>
@@ -852,7 +852,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>199.46</td>
+<td>0.20</td>
 <td>109.80</td>
 <td>107.97</td>
 <td>1628.87</td>
@@ -865,7 +865,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">seal_recognition-nopp-nolayout</td>
 <td>Intel 6271C</td>
-<td>662.30</td>
+<td>0.66</td>
 <td>1021.40</td>
 <td>1014.05</td>
 <td>2541.84</td>
@@ -877,7 +877,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>765.66</td>
+<td>0.77</td>
 <td>1013.70</td>
 <td>1009.71</td>
 <td>2558.11</td>
@@ -889,7 +889,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>102.32</td>
+<td>0.10</td>
 <td>113.90</td>
 <td>112.78</td>
 <td>1556.05</td>
@@ -901,7 +901,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>63.93</td>
+<td>0.06</td>
 <td>122.90</td>
 <td>120.53</td>
 <td>1323.04</td>
@@ -913,7 +913,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>97.27</td>
+<td>0.10</td>
 <td>123.80</td>
 <td>121.95</td>
 <td>1349.10</td>
@@ -925,7 +925,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>97.39</td>
+<td>0.10</td>
 <td>116.90</td>
 <td>113.28</td>
 <td>1620.22</td>
@@ -937,7 +937,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>81.86</td>
+<td>0.08</td>
 <td>123.50</td>
 <td>120.70</td>
 <td>1477.95</td>
@@ -949,7 +949,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>2045.01</td>
+<td>2.05</td>
 <td>103.60</td>
 <td>101.33</td>
 <td>4115.47</td>
@@ -961,7 +961,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>174.25</td>
+<td>0.17</td>
 <td>112.10</td>
 <td>109.61</td>
 <td>1477.02</td>
@@ -974,7 +974,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">seal_recognition-nopp-lightweight</td>
 <td>Intel 6271C</td>
-<td>647.56</td>
+<td>0.65</td>
 <td>1015.50</td>
 <td>1010.90</td>
 <td>1701.61</td>
@@ -986,7 +986,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>600.89</td>
+<td>0.60</td>
 <td>1011.70</td>
 <td>1004.39</td>
 <td>1794.69</td>
@@ -998,7 +998,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>71.31</td>
+<td>0.07</td>
 <td>120.90</td>
 <td>118.70</td>
 <td>1601.37</td>
@@ -1010,7 +1010,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>61.25</td>
+<td>0.06</td>
 <td>123.90</td>
 <td>121.47</td>
 <td>1483.47</td>
@@ -1022,7 +1022,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>90.24</td>
+<td>0.09</td>
 <td>125.80</td>
 <td>124.22</td>
 <td>1410.70</td>
@@ -1034,7 +1034,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>49.11</td>
+<td>0.05</td>
 <td>131.90</td>
 <td>124.07</td>
 <td>1820.23</td>
@@ -1046,7 +1046,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>62.53</td>
+<td>0.06</td>
 <td>131.80</td>
 <td>129.06</td>
 <td>1668.08</td>
@@ -1058,7 +1058,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>696.42</td>
+<td>0.70</td>
 <td>106.60</td>
 <td>104.71</td>
 <td>1176.22</td>
@@ -1070,7 +1070,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>89.92</td>
+<td>0.09</td>
 <td>120.80</td>
 <td>117.79</td>
 <td>1492.52</td>
@@ -1083,7 +1083,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">seal_recognition-nopp-lightweightlayout</td>
 <td>Intel 6271C</td>
-<td>256.55</td>
+<td>0.26</td>
 <td>1018.90</td>
 <td>1012.30</td>
 <td>2759.20</td>
@@ -1095,7 +1095,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>294.96</td>
+<td>0.29</td>
 <td>1017.60</td>
 <td>1009.56</td>
 <td>2718.50</td>
@@ -1107,7 +1107,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>55.61</td>
+<td>0.06</td>
 <td>136.30</td>
 <td>128.47</td>
 <td>1565.15</td>
@@ -1119,7 +1119,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>39.48</td>
+<td>0.04</td>
 <td>115.90</td>
 <td>115.12</td>
 <td>1313.98</td>
@@ -1131,7 +1131,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>60.79</td>
+<td>0.06</td>
 <td>119.80</td>
 <td>117.14</td>
 <td>1268.04</td>
@@ -1143,7 +1143,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>48.73</td>
+<td>0.05</td>
 <td>119.00</td>
 <td>111.67</td>
 <td>1735.01</td>
@@ -1155,7 +1155,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>46.02</td>
+<td>0.05</td>
 <td>118.90</td>
 <td>118.48</td>
 <td>1596.27</td>
@@ -1167,7 +1167,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>789.23</td>
+<td>0.79</td>
 <td>112.40</td>
 <td>102.80</td>
 <td>6106.27</td>
@@ -1179,7 +1179,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>79.36</td>
+<td>0.08</td>
 <td>111.90</td>
 <td>108.66</td>
 <td>1510.57</td>
@@ -1192,7 +1192,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">seal_recognition-nopp-nolayout-lightweight</td>
 <td>Intel 6271C</td>
-<td>149.97</td>
+<td>0.15</td>
 <td>1027.50</td>
 <td>1022.43</td>
 <td>1245.17</td>
@@ -1204,7 +1204,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>126.58</td>
+<td>0.13</td>
 <td>1024.70</td>
 <td>1014.78</td>
 <td>1211.13</td>
@@ -1216,7 +1216,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>57.79</td>
+<td>0.06</td>
 <td>131.70</td>
 <td>128.79</td>
 <td>1532.17</td>
@@ -1228,7 +1228,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>42.42</td>
+<td>0.04</td>
 <td>133.00</td>
 <td>129.92</td>
 <td>1169.25</td>
@@ -1240,7 +1240,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>68.18</td>
+<td>0.07</td>
 <td>132.70</td>
 <td>128.90</td>
 <td>1269.38</td>
@@ -1252,7 +1252,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>39.22</td>
+<td>0.04</td>
 <td>134.90</td>
 <td>128.48</td>
 <td>1530.35</td>
@@ -1264,7 +1264,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>47.19</td>
+<td>0.05</td>
 <td>141.90</td>
 <td>139.48</td>
 <td>1458.07</td>
@@ -1276,7 +1276,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>275.15</td>
+<td>0.28</td>
 <td>110.30</td>
 <td>109.19</td>
 <td>1100.36</td>
@@ -1288,7 +1288,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>68.63</td>
+<td>0.07</td>
 <td>125.90</td>
 <td>123.15</td>
 <td>1389.43</td>
@@ -1335,7 +1335,10 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing 130 seal images.
+* Test strategy:
+    * Warm up with 20 samples, then repeat the full dataset once for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.md

@@ -662,11 +662,11 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="9">seal_recognition-default</td>
 <td>Intel 6271C</td>
-<td>1441.99</td>
+<td>1.44</td>
 <td>1013.50</td>
 <td>1005.77</td>
 <td>2850.27</td>
@@ -678,7 +678,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1303.84</td>
+<td>1.30</td>
 <td>1010.40</td>
 <td>970.29</td>
 <td>2797.28</td>
@@ -690,7 +690,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>102.05</td>
+<td>0.10</td>
 <td>115.80</td>
 <td>112.72</td>
 <td>1647.39</td>
@@ -702,7 +702,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>67.80</td>
+<td>0.07</td>
 <td>117.90</td>
 <td>115.03</td>
 <td>1519.09</td>
@@ -714,7 +714,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>115.01</td>
+<td>0.12</td>
 <td>117.80</td>
 <td>115.16</td>
 <td>1425.80</td>
@@ -726,7 +726,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>77.40</td>
+<td>0.08</td>
 <td>117.40</td>
 <td>111.64</td>
 <td>1721.26</td>
@@ -738,7 +738,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>79.64</td>
+<td>0.08</td>
 <td>125.80</td>
 <td>118.96</td>
 <td>1727.16</td>
@@ -750,7 +750,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>1543.47</td>
+<td>1.54</td>
 <td>104.70</td>
 <td>101.57</td>
 <td>4526.11</td>
@@ -762,7 +762,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>155.57</td>
+<td>0.16</td>
 <td>110.20</td>
 <td>108.01</td>
 <td>1690.79</td>
@@ -775,7 +775,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">seal_recognition-nopp</td>
 <td>Intel 6271C</td>
-<td>1109.72</td>
+<td>1.11</td>
 <td>1014.70</td>
 <td>1009.28</td>
 <td>3062.18</td>
@@ -787,7 +787,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>1203.15</td>
+<td>1.20</td>
 <td>1012.60</td>
 <td>1006.68</td>
 <td>3079.39</td>
@@ -799,7 +799,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>119.58</td>
+<td>0.12</td>
 <td>116.90</td>
 <td>115.51</td>
 <td>1608.80</td>
@@ -811,7 +811,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>76.98</td>
+<td>0.08</td>
 <td>118.90</td>
 <td>117.49</td>
 <td>1440.84</td>
@@ -823,7 +823,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>124.81</td>
+<td>0.12</td>
 <td>119.80</td>
 <td>117.69</td>
 <td>1451.77</td>
@@ -835,7 +835,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>102.76</td>
+<td>0.10</td>
 <td>117.90</td>
 <td>111.53</td>
 <td>1722.02</td>
@@ -847,7 +847,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>95.30</td>
+<td>0.10</td>
 <td>120.80</td>
 <td>118.38</td>
 <td>1642.00</td>
@@ -859,7 +859,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>2343.35</td>
+<td>2.34</td>
 <td>104.30</td>
 <td>101.47</td>
 <td>4436.23</td>
@@ -871,7 +871,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>199.46</td>
+<td>0.20</td>
 <td>109.80</td>
 <td>107.97</td>
 <td>1628.87</td>
@@ -884,7 +884,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">seal_recognition-nopp-nolayout</td>
 <td>Intel 6271C</td>
-<td>662.30</td>
+<td>0.66</td>
 <td>1021.40</td>
 <td>1014.05</td>
 <td>2541.84</td>
@@ -896,7 +896,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>765.66</td>
+<td>0.77</td>
 <td>1013.70</td>
 <td>1009.71</td>
 <td>2558.11</td>
@@ -908,7 +908,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>102.32</td>
+<td>0.10</td>
 <td>113.90</td>
 <td>112.78</td>
 <td>1556.05</td>
@@ -920,7 +920,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>63.93</td>
+<td>0.06</td>
 <td>122.90</td>
 <td>120.53</td>
 <td>1323.04</td>
@@ -932,7 +932,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>97.27</td>
+<td>0.10</td>
 <td>123.80</td>
 <td>121.95</td>
 <td>1349.10</td>
@@ -944,7 +944,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>97.39</td>
+<td>0.10</td>
 <td>116.90</td>
 <td>113.28</td>
 <td>1620.22</td>
@@ -956,7 +956,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>81.86</td>
+<td>0.08</td>
 <td>123.50</td>
 <td>120.70</td>
 <td>1477.95</td>
@@ -968,7 +968,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>2045.01</td>
+<td>2.05</td>
 <td>103.60</td>
 <td>101.33</td>
 <td>4115.47</td>
@@ -980,7 +980,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>174.25</td>
+<td>0.17</td>
 <td>112.10</td>
 <td>109.61</td>
 <td>1477.02</td>
@@ -993,7 +993,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">seal_recognition-nopp-lightweight</td>
 <td>Intel 6271C</td>
-<td>647.56</td>
+<td>0.65</td>
 <td>1015.50</td>
 <td>1010.90</td>
 <td>1701.61</td>
@@ -1005,7 +1005,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>600.89</td>
+<td>0.60</td>
 <td>1011.70</td>
 <td>1004.39</td>
 <td>1794.69</td>
@@ -1017,7 +1017,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>71.31</td>
+<td>0.07</td>
 <td>120.90</td>
 <td>118.70</td>
 <td>1601.37</td>
@@ -1029,7 +1029,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>61.25</td>
+<td>0.06</td>
 <td>123.90</td>
 <td>121.47</td>
 <td>1483.47</td>
@@ -1041,7 +1041,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>90.24</td>
+<td>0.09</td>
 <td>125.80</td>
 <td>124.22</td>
 <td>1410.70</td>
@@ -1053,7 +1053,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>49.11</td>
+<td>0.05</td>
 <td>131.90</td>
 <td>124.07</td>
 <td>1820.23</td>
@@ -1065,7 +1065,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>62.53</td>
+<td>0.06</td>
 <td>131.80</td>
 <td>129.06</td>
 <td>1668.08</td>
@@ -1077,7 +1077,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>696.42</td>
+<td>0.70</td>
 <td>106.60</td>
 <td>104.71</td>
 <td>1176.22</td>
@@ -1089,7 +1089,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>89.92</td>
+<td>0.09</td>
 <td>120.80</td>
 <td>117.79</td>
 <td>1492.52</td>
@@ -1102,7 +1102,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">seal_recognition-nopp-lightweightlayout</td>
 <td>Intel 6271C</td>
-<td>256.55</td>
+<td>0.26</td>
 <td>1018.90</td>
 <td>1012.30</td>
 <td>2759.20</td>
@@ -1114,7 +1114,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>294.96</td>
+<td>0.29</td>
 <td>1017.60</td>
 <td>1009.56</td>
 <td>2718.50</td>
@@ -1126,7 +1126,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>55.61</td>
+<td>0.06</td>
 <td>136.30</td>
 <td>128.47</td>
 <td>1565.15</td>
@@ -1138,7 +1138,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>39.48</td>
+<td>0.04</td>
 <td>115.90</td>
 <td>115.12</td>
 <td>1313.98</td>
@@ -1150,7 +1150,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>60.79</td>
+<td>0.06</td>
 <td>119.80</td>
 <td>117.14</td>
 <td>1268.04</td>
@@ -1162,7 +1162,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>48.73</td>
+<td>0.05</td>
 <td>119.00</td>
 <td>111.67</td>
 <td>1735.01</td>
@@ -1174,7 +1174,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>46.02</td>
+<td>0.05</td>
 <td>118.90</td>
 <td>118.48</td>
 <td>1596.27</td>
@@ -1186,7 +1186,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>789.23</td>
+<td>0.79</td>
 <td>112.40</td>
 <td>102.80</td>
 <td>6106.27</td>
@@ -1198,7 +1198,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>79.36</td>
+<td>0.08</td>
 <td>111.90</td>
 <td>108.66</td>
 <td>1510.57</td>
@@ -1211,7 +1211,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">seal_recognition-nopp-nolayout-lightweight</td>
 <td>Intel 6271C</td>
-<td>149.97</td>
+<td>0.15</td>
 <td>1027.50</td>
 <td>1022.43</td>
 <td>1245.17</td>
@@ -1223,7 +1223,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>126.58</td>
+<td>0.13</td>
 <td>1024.70</td>
 <td>1014.78</td>
 <td>1211.13</td>
@@ -1235,7 +1235,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>57.79</td>
+<td>0.06</td>
 <td>131.70</td>
 <td>128.79</td>
 <td>1532.17</td>
@@ -1247,7 +1247,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>42.42</td>
+<td>0.04</td>
 <td>133.00</td>
 <td>129.92</td>
 <td>1169.25</td>
@@ -1259,7 +1259,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>68.18</td>
+<td>0.07</td>
 <td>132.70</td>
 <td>128.90</td>
 <td>1269.38</td>
@@ -1271,7 +1271,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>39.22</td>
+<td>0.04</td>
 <td>134.90</td>
 <td>128.48</td>
 <td>1530.35</td>
@@ -1283,7 +1283,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>47.19</td>
+<td>0.05</td>
 <td>141.90</td>
 <td>139.48</td>
 <td>1458.07</td>
@@ -1295,7 +1295,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>275.15</td>
+<td>0.28</td>
 <td>110.30</td>
 <td>109.19</td>
 <td>1100.36</td>
@@ -1307,7 +1307,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>68.63</td>
+<td>0.07</td>
 <td>125.90</td>
 <td>123.15</td>
 <td>1389.43</td>
@@ -1354,7 +1354,10 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含130张印章图像。
+* 测试策略：
+    * 使用 20 个样本进行预热，然后对整个数据集重复 1 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
 

docs/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.en.md

@@ -779,11 +779,11 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
 <td rowspan="7">table_recognition_v2-default</td>
 <td>Intel 6271C</td>
-<td>4682.50</td>
+<td>4.68</td>
 <td>1150.80</td>
 <td>1005.33</td>
 <td>4391.37</td>
@@ -795,7 +795,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>4737.37</td>
+<td>4.74</td>
 <td>1009.60</td>
 <td>945.16</td>
 <td>4370.65</td>
@@ -807,7 +807,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>683.64</td>
+<td>0.68</td>
 <td>120.40</td>
 <td>107.35</td>
 <td>1936.46</td>
@@ -819,7 +819,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>785.91</td>
+<td>0.79</td>
 <td>246.70</td>
 <td>137.46</td>
 <td>1814.86</td>
@@ -831,7 +831,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>512.60</td>
+<td>0.51</td>
 <td>533.50</td>
 <td>260.91</td>
 <td>2544.58</td>
@@ -843,7 +843,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>579.97</td>
+<td>0.58</td>
 <td>131.80</td>
 <td>4.18</td>
 <td>2044.72</td>
@@ -855,7 +855,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>4316.55</td>
+<td>4.32</td>
 <td>226.00</td>
 <td>103.48</td>
 <td>4951.52</td>
@@ -868,7 +868,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="8">table_recognition_v2-nopp</td>
 <td>Intel 6271C</td>
-<td>6124.28</td>
+<td>6.12</td>
 <td>1027.40</td>
 <td>1001.08</td>
 <td>4333.77</td>
@@ -880,7 +880,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>5976.96</td>
+<td>5.98</td>
 <td>1010.70</td>
 <td>940.00</td>
 <td>4297.76</td>
@@ -892,7 +892,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>778.48</td>
+<td>0.78</td>
 <td>109.90</td>
 <td>104.26</td>
 <td>1975.51</td>
@@ -904,7 +904,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>915.91</td>
+<td>0.92</td>
 <td>107.80</td>
 <td>102.97</td>
 <td>1725.44</td>
@@ -916,7 +916,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>568.28</td>
+<td>0.57</td>
 <td>107.80</td>
 <td>101.55</td>
 <td>2115.62</td>
@@ -928,7 +928,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>665.59</td>
+<td>0.67</td>
 <td>105.90</td>
 <td>102.23</td>
 <td>1966.66</td>
@@ -940,7 +940,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>5354.34</td>
+<td>5.35</td>
 <td>106.70</td>
 <td>103.34</td>
 <td>4906.53</td>
@@ -952,7 +952,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>970.53</td>
+<td>0.97</td>
 <td>105.30</td>
 <td>101.99</td>
 <td>1945.11</td>
@@ -965,7 +965,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 <tr>
 <td rowspan="9">table_recognition_v2-slanet</td>
 <td>Intel 6271C</td>
-<td>3828.52</td>
+<td>3.83</td>
 <td>1178.40</td>
 <td>1013.64</td>
 <td>3617.43</td>
@@ -977,7 +977,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3800.14</td>
+<td>3.80</td>
 <td>1034.60</td>
 <td>986.05</td>
 <td>3629.65</td>
@@ -989,7 +989,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>675.10</td>
+<td>0.68</td>
 <td>120.90</td>
 <td>107.85</td>
 <td>1747.88</td>
@@ -1001,7 +1001,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>544.97</td>
+<td>0.54</td>
 <td>374.60</td>
 <td>153.89</td>
 <td>1900.24</td>
@@ -1013,7 +1013,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>774.21</td>
+<td>0.77</td>
 <td>253.70</td>
 <td>139.34</td>
 <td>1636.04</td>
@@ -1025,7 +1025,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>503.17</td>
+<td>0.50</td>
 <td>963.00</td>
 <td>427.93</td>
 <td>2051.32</td>
@@ -1037,7 +1037,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>567.63</td>
+<td>0.57</td>
 <td>135.20</td>
 <td>108.01</td>
 <td>1930.06</td>
@@ -1049,7 +1049,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>M4</td>
-<td>5562.14</td>
+<td>5.56</td>
 <td>223.10</td>
 <td>103.73</td>
 <td>4456.39</td>
@@ -1061,7 +1061,7 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>784.60</td>
+<td>0.78</td>
 <td>412.90</td>
 <td>178.49</td>
 <td>1839.10</td>
@@ -1096,7 +1096,10 @@ The ultra-lightweight cyrillic alphabet recognition model trained based on the P
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing 5 images of various categories, such as large tables, small tables, bordered tables, and borderless tables.
+* Test strategy:
+    * Warm up with 3 samples, then repeat the full dataset 20 times for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.md

@@ -797,11 +797,11 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="7">table_recognition_v2-default</td>
 <td>Intel 6271C</td>
-<td>4682.50</td>
+<td>4.68</td>
 <td>1150.80</td>
 <td>1005.33</td>
 <td>4391.37</td>
@@ -813,7 +813,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>4737.37</td>
+<td>4.74</td>
 <td>1009.60</td>
 <td>945.16</td>
 <td>4370.65</td>
@@ -825,7 +825,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>683.64</td>
+<td>0.68</td>
 <td>120.40</td>
 <td>107.35</td>
 <td>1936.46</td>
@@ -837,7 +837,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>785.91</td>
+<td>0.79</td>
 <td>246.70</td>
 <td>137.46</td>
 <td>1814.86</td>
@@ -849,7 +849,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>512.60</td>
+<td>0.51</td>
 <td>533.50</td>
 <td>260.91</td>
 <td>2544.58</td>
@@ -861,7 +861,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>579.97</td>
+<td>0.58</td>
 <td>131.80</td>
 <td>4.18</td>
 <td>2044.72</td>
@@ -873,7 +873,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>4316.55</td>
+<td>4.32</td>
 <td>226.00</td>
 <td>103.48</td>
 <td>4951.52</td>
@@ -886,7 +886,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="8">table_recognition_v2-nopp</td>
 <td>Intel 6271C</td>
-<td>6124.28</td>
+<td>6.12</td>
 <td>1027.40</td>
 <td>1001.08</td>
 <td>4333.77</td>
@@ -898,7 +898,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>5976.96</td>
+<td>5.98</td>
 <td>1010.70</td>
 <td>940.00</td>
 <td>4297.76</td>
@@ -910,7 +910,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>778.48</td>
+<td>0.78</td>
 <td>109.90</td>
 <td>104.26</td>
 <td>1975.51</td>
@@ -922,7 +922,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>915.91</td>
+<td>0.92</td>
 <td>107.80</td>
 <td>102.97</td>
 <td>1725.44</td>
@@ -934,7 +934,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>568.28</td>
+<td>0.57</td>
 <td>107.80</td>
 <td>101.55</td>
 <td>2115.62</td>
@@ -946,7 +946,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>665.59</td>
+<td>0.67</td>
 <td>105.90</td>
 <td>102.23</td>
 <td>1966.66</td>
@@ -958,7 +958,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>5354.34</td>
+<td>5.35</td>
 <td>106.70</td>
 <td>103.34</td>
 <td>4906.53</td>
@@ -970,7 +970,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>970.53</td>
+<td>0.97</td>
 <td>105.30</td>
 <td>101.99</td>
 <td>1945.11</td>
@@ -983,7 +983,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 <tr>
 <td rowspan="9">table_recognition_v2-slanet</td>
 <td>Intel 6271C</td>
-<td>3828.52</td>
+<td>3.83</td>
 <td>1178.40</td>
 <td>1013.64</td>
 <td>3617.43</td>
@@ -995,7 +995,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C</td>
-<td>3800.14</td>
+<td>3.80</td>
 <td>1034.60</td>
 <td>986.05</td>
 <td>3629.65</td>
@@ -1007,7 +1007,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Hygon 7490 + P800</td>
-<td>675.10</td>
+<td>0.68</td>
 <td>120.90</td>
 <td>107.85</td>
 <td>1747.88</td>
@@ -1019,7 +1019,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A100</td>
-<td>544.97</td>
+<td>0.54</td>
 <td>374.60</td>
 <td>153.89</td>
 <td>1900.24</td>
@@ -1031,7 +1031,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + V100</td>
-<td>774.21</td>
+<td>0.77</td>
 <td>253.70</td>
 <td>139.34</td>
 <td>1636.04</td>
@@ -1043,7 +1043,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8563C + H20</td>
-<td>503.17</td>
+<td>0.50</td>
 <td>963.00</td>
 <td>427.93</td>
 <td>2051.32</td>
@@ -1055,7 +1055,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>567.63</td>
+<td>0.57</td>
 <td>135.20</td>
 <td>108.01</td>
 <td>1930.06</td>
@@ -1067,7 +1067,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>M4</td>
-<td>5562.14</td>
+<td>5.56</td>
 <td>223.10</td>
 <td>103.73</td>
 <td>4456.39</td>
@@ -1079,7 +1079,7 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
 </tr>
 <tr>
 <td>Intel 6271C + T4</td>
-<td>784.60</td>
+<td>0.78</td>
 <td>412.90</td>
 <td>178.49</td>
 <td>1839.10</td>
@@ -1114,7 +1114,10 @@ devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://padd
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含大尺寸表格、小尺寸表格、有线表、无线表等多类别的5张图像。
+* 测试策略：
+    * 使用 3 个样本进行预热，然后对整个数据集重复 20 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
 

docs/pipeline_usage/tutorials/vlm_pipelines/doc_understanding.en.md

@@ -50,333 +50,31 @@ The Document Understanding Pipeline is an advanced document processing technolog
 <summary>Click to expand/collapse the table</summary>
 
 <table border="1">
-<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (ms)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
-<tr>
-<td rowspan="9">doc_preprocessor-default</td>
-<td>Intel 6271C</td>
-<td>1127.85</td>
-<td>1012.50</td>
-<td>789.07</td>
-<td>1913.89</td>
-<td>1650.30</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 8350C</td>
-<td>764.23</td>
-<td>1003.70</td>
-<td>782.77</td>
-<td>1916.81</td>
-<td>1655.66</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Hygon 7490 + P800</td>
-<td>96.74</td>
-<td>174.70</td>
-<td>135.91</td>
-<td>2023.54</td>
-<td>1963.53</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 8350C + A100</td>
-<td>76.66</td>
-<td>148.90</td>
-<td>125.73</td>
-<td>1748.06</td>
-<td>1683.35</td>
-<td>11</td>
-<td>5.86</td>
-<td>828.00</td>
-<td>828.00</td>
-</tr>
-<tr>
-<td>Intel 6271C + V100</td>
-<td>114.96</td>
-<td>168.60</td>
-<td>131.05</td>
-<td>1837.66</td>
-<td>1707.50</td>
-<td>26</td>
-<td>11.65</td>
-<td>694.00</td>
-<td>694.00</td>
-</tr>
-<tr>
-<td>Intel 8563C + H20</td>
-<td>69.34</td>
-<td>140.90</td>
-<td>122.80</td>
-<td>1992.86</td>
-<td>1911.34</td>
-<td>11</td>
-<td>7.75</td>
-<td>890.00</td>
-<td>890.00</td>
-</tr>
-<tr>
-<td>Intel 8350C + A10</td>
-<td>81.65</td>
-<td>146.40</td>
-<td>125.37</td>
-<td>1896.46</td>
-<td>1825.98</td>
-<td>14</td>
-<td>5.93</td>
-<td>586.00</td>
-<td>586.00</td>
-</tr>
-<tr>
-<td>M4</td>
-<td>368.93</td>
-<td>118.20</td>
-<td>103.40</td>
-<td>1973.23</td>
-<td>1756.00</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 6271C + T4</td>
-<td>122.40</td>
-<td>156.90</td>
-<td>123.75</td>
-<td>1874.29</td>
-<td>1727.01</td>
-<td>73</td>
-<td>23.81</td>
-<td>450.00</td>
-<td>450.00</td>
-</tr>
-<tr>
-<td rowspan="9">doc_preprocessor-clsonly</td>
-<td>Intel 6271C</td>
-<td>55.12</td>
-<td>1085.60</td>
-<td>1057.12</td>
-<td>1254.27</td>
-<td>1203.36</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 8350C</td>
-<td>59.48</td>
-<td>1051.60</td>
-<td>1043.39</td>
-<td>1263.88</td>
-<td>1206.67</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
 <tr>
+<td rowspan="2">doc_understanding-default</td>
 <td>Hygon 7490 + P800</td>
-<td>47.44</td>
-<td>202.70</td>
-<td>173.75</td>
-<td>1868.11</td>
-<td>1810.80</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 8350C + A100</td>
-<td>43.27</td>
-<td>167.80</td>
-<td>150.56</td>
-<td>1347.92</td>
-<td>1325.32</td>
-<td>2</td>
-<td>1.25</td>
-<td>514.00</td>
-<td>514.00</td>
-</tr>
-<tr>
-<td>Intel 6271C + V100</td>
-<td>53.72</td>
-<td>202.70</td>
-<td>170.72</td>
-<td>1404.16</td>
-<td>1352.74</td>
-<td>2</td>
-<td>1.20</td>
-<td>394.00</td>
-<td>394.00</td>
-</tr>
-<tr>
-<td>Intel 8563C + H20</td>
-<td>41.89</td>
-<td>153.80</td>
-<td>141.46</td>
-<td>1590.48</td>
-<td>1540.25</td>
-<td>2</td>
-<td>1.12</td>
-<td>578.00</td>
-<td>578.00</td>
-</tr>
-<tr>
-<td>Intel 8350C + A10</td>
-<td>44.61</td>
-<td>179.80</td>
-<td>156.96</td>
-<td>1569.85</td>
-<td>1514.64</td>
-<td>2</td>
-<td>0.88</td>
-<td>302.00</td>
-<td>302.00</td>
-</tr>
-<tr>
-<td>M4</td>
-<td>34.12</td>
-<td>127.60</td>
-<td>123.24</td>
-<td>1459.06</td>
-<td>1386.33</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 6271C + T4</td>
-<td>51.81</td>
-<td>180.80</td>
-<td>154.34</td>
-<td>1515.43</td>
-<td>1465.99</td>
-<td>4</td>
-<td>2.89</td>
-<td>160.00</td>
-<td>160.00</td>
-</tr>
-<tr>
-<td rowspan="9">doc_preprocessor-unwarponly</td>
-<td>Intel 6271C</td>
-<td>1092.82</td>
-<td>1002.10</td>
-<td>749.82</td>
-<td>1875.45</td>
-<td>1680.20</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 8350C</td>
-<td>747.32</td>
-<td>951.30</td>
-<td>754.01</td>
-<td>1843.54</td>
-<td>1645.73</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Hygon 7490 + P800</td>
-<td>95.22</td>
-<td>101.20</td>
-<td>100.08</td>
-<td>1960.48</td>
-<td>1863.72</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-<td>N/A</td>
-</tr>
-<tr>
-<td>Intel 8350C + A100</td>
-<td>78.05</td>
-<td>100.90</td>
-<td>100.15</td>
-<td>1570.25</td>
-<td>1446.01</td>
-<td>9</td>
-<td>5.57</td>
-<td>788.00</td>
-<td>788.00</td>
-</tr>
-<tr>
-<td>Intel 6271C + V100</td>
-<td>130.84</td>
-<td>100.90</td>
-<td>100.25</td>
-<td>1561.52</td>
-<td>1459.73</td>
-<td>24</td>
-<td>11.35</td>
-<td>656.00</td>
-<td>656.00</td>
-</tr>
-<tr>
-<td>Intel 8563C + H20</td>
-<td>68.62</td>
-<td>106.90</td>
-<td>100.62</td>
-<td>1808.00</td>
-<td>1715.35</td>
-<td>6</td>
-<td>4.83</td>
-<td>684.00</td>
-<td>684.00</td>
-</tr>
-<tr>
-<td>Intel 8350C + A10</td>
-<td>84.19</td>
+<td>26.61</td>
 <td>101.90</td>
-<td>100.14</td>
-<td>1848.25</td>
-<td>1719.24</td>
-<td>19</td>
-<td>7.73</td>
-<td>544.00</td>
-<td>544.00</td>
-</tr>
-<tr>
-<td>M4</td>
-<td>295.15</td>
-<td>102.40</td>
-<td>100.76</td>
-<td>1957.42</td>
-<td>1734.67</td>
+<td>100.07</td>
+<td>3112.36</td>
+<td>3112.27</td>
 <td>N/A</td>
 <td>N/A</td>
 <td>N/A</td>
 <td>N/A</td>
 </tr>
 <tr>
-<td>Intel 6271C + T4</td>
-<td>117.25</td>
-<td>101.80</td>
-<td>100.36</td>
-<td>1674.25</td>
-<td>1605.14</td>
-<td>48</td>
-<td>26.45</td>
-<td>412.00</td>
-<td>412.00</td>
+<td>Intel 8350C + A10</td>
+<td>16.89</td>
+<td>101.70</td>
+<td>100.09</td>
+<td>2959.31</td>
+<td>2959.25</td>
+<td>100</td>
+<td>42.85</td>
+<td>17254.00</td>
+<td>17252.01</td>
 </tr>
 </table>
 
@@ -384,17 +82,9 @@ The Document Understanding Pipeline is an advanced document processing technolog
 <table border="1">
 <tr><th>Pipeline configuration</th><th>description</th></tr>
 <tr>
-<td>doc_preprocessor-default</td>
+<td>doc_understanding-default</td>
 <td>Default configuration</td>
 </tr>
-<tr>
-<td>doc_preprocessor-clsonly</td>
-<td>Based on the default configuration, only document image orientation classification is enabled</td>
-</tr>
-<tr>
-<td>doc_preprocessor-unwarponly</td>
-<td>Based on the default configuration, only text image rectification is enabled</td>
-</tr>
 </table>
 </details>
 
@@ -403,7 +93,10 @@ The Document Understanding Pipeline is an advanced document processing technolog
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* Test data:
+    * Test data containing one document parsing sample and one chatQA sample, each sample contains both image and JSON formats.
+* Test strategy:
+    * Warm up with 2 samples, then repeat the full dataset 20 times for performance testing.
 * Note:
     * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
 

docs/pipeline_usage/tutorials/vlm_pipelines/doc_understanding.md

@@ -53,11 +53,11 @@ comments: true
 <summary>点击展开/折叠表格</summary>
 
 <table border="1">
-<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (ms)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
 <tr>
 <td rowspan="2">doc_understanding-default</td>
 <td>Hygon 7490 + P800</td>
-<td>26614.69</td>
+<td>26.61</td>
 <td>101.90</td>
 <td>100.07</td>
 <td>3112.36</td>
@@ -69,7 +69,7 @@ comments: true
 </tr>
 <tr>
 <td>Intel 8350C + A10</td>
-<td>16887.80</td>
+<td>16.89</td>
 <td>101.70</td>
 <td>100.09</td>
 <td>2959.31</td>
@@ -96,7 +96,10 @@ comments: true
     * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
     * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
     * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
-
+* 测试数据：
+    * 测试数据包含一个文档解析和一个chatQA的样本，每个样本包含图像、json两种形式。
+* 测试策略：
+    * 使用 2 个样本进行预热，然后对整个数据集重复 20 次以进行速度性能测试。
 * 备注：
     * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。