add 3cls model (#2020)

* add 3cls model

* add 3cls model

* cls_config_fix

* cls_config_fix

* fix seg convert bug

* cls oom

* Update MobileNetV4_hybrid_large.yaml

* Update MobileNetV4_hybrid_medium.yaml

* rename model name and official model

README.md

@@ -63,7 +63,7 @@ PaddleX 3.0 覆盖了 16 条产业级模型产线，其中 9 条基础产线可
     <td rowspan="2">通用图像分类</td>
     <td>图像分类</td>
     <td>CLIP_vit_base_patch16_224<br/>CLIP_vit_large_patch14_224<details>
-    <summary><b>more</b></summary><br/>ConvNeXt_tiny<br/>ConvNeXt_small<br/>ConvNeXt_base_224<br/>ConvNeXt_base_384<br/>ConvNeXt_large_224<br/>ConvNeXt_large_384<br/>MobileNetV1_x0_25<br/>MobileNetV1_x0_5<br/>MobileNetV1_x0_75<br/>MobileNetV1_x1_0<br/>MobileNetV2_x0_25<br/>MobileNetV2_x0_5<br/>MobileNetV2_x1_0<br/>MobileNetV2_x1_5<br/>MobileNetV2_x2_0<br/>MobileNetV3_large_x0_35<br/>MobileNetV3_large_x0_5<br/>MobileNetV3_large_x0_75<br/>MobileNetV3_large_x1_0<br/>MobileNetV3_large_x1_25<br/>MobileNetV3_small_x0_35<br/>MobileNetV3_small_x0_5<br/>MobileNetV3_small_x0_75<br/>MobileNetV3_small_x1_0<br/>MobileNetV3_small_x1_25<br/>PP-HGNet_tiny<br/>PP-HGNet_small<br/>PP-HGNet_base<br/>PP-HGNetV2-B0<br/>PP-HGNetV2-B1<br/>PP-HGNetV2-B2<br/>PP-HGNetV2-B3<br/>PP-HGNetV2-B4<br/>PP-HGNetV2-B5<br/>PP-HGNetV2-B6<br/>PP-LCNet_x0_25<br/>PP-LCNet_x0_35<br/>PP-LCNet_x0_5<br/>PP-LCNet_x0_75<br/>PP-LCNet_x1_0<br/>PP-LCNet_x1_5<br/>PP-LCNet_x2_0<br/>PP-LCNet_x2_5<br/>PP-LCNetV2_small<br/>PP-LCNetV2_base<br/>PP-LCNetV2_large<br/>ResNet18<br/>ResNet18_vd<br/>ResNet34<br/>ResNet34_vd<br/>ResNet50<br/>ResNet50_vd<br/>ResNet101<br/>ResNet101_vd<br/>ResNet152<br/>ResNet152_vd<br/>ResNet200_vd<br/>SwinTransformer_tiny_patch4_window7_224<br/>SwinTransformer_small_patch4_window7_224<br/>SwinTransformer_base_patch4_window7_224<br/>SwinTransformer_base_patch4_window12_384<br/>SwinTransformer_large_patch4_window7_224<br/>SwinTransformer_large_patch4_window12_384</details></td>
+    <summary><b>more</b></summary><br/>ConvNeXt_tiny<br/>ConvNeXt_small<br/>ConvNeXt_base_224<br/>ConvNeXt_base_384<br/>ConvNeXt_large_224<br/>ConvNeXt_large_384<br/>MobileNetV1_x0_25<br/>MobileNetV1_x0_5<br/>MobileNetV1_x0_75<br/>MobileNetV1_x1_0<br/>MobileNetV2_x0_25<br/>MobileNetV2_x0_5<br/>MobileNetV2_x1_0<br/>MobileNetV2_x1_5<br/>MobileNetV2_x2_0<br/>MobileNetV3_large_x0_35<br/>MobileNetV3_large_x0_5<br/>MobileNetV3_large_x0_75<br/>MobileNetV3_large_x1_0<br/>MobileNetV3_large_x1_25<br/>MobileNetV3_small_x0_35<br/>MobileNetV3_small_x0_5<br/>MobileNetV3_small_x0_75<br/>MobileNetV3_small_x1_0<br/>MobileNetV3_small_x1_25<br/>MobileNetV4_conv_small<br/>MobileNetV4_conv_medium<br/>MobileNetV4_conv_large<br/>MobileNetV4_hybrid_medium<br/>MobileNetV4_hybrid_large<br/>PP-HGNet_tiny<br/>PP-HGNet_small<br/>PP-HGNet_base<br/>PP-HGNetV2-B0<br/>PP-HGNetV2-B1<br/>PP-HGNetV2-B2<br/>PP-HGNetV2-B3<br/>PP-HGNetV2-B4<br/>PP-HGNetV2-B5<br/>PP-HGNetV2-B6<br/>PP-LCNet_x0_25<br/>PP-LCNet_x0_35<br/>PP-LCNet_x0_5<br/>PP-LCNet_x0_75<br/>PP-LCNet_x1_0<br/>PP-LCNet_x1_5<br/>PP-LCNet_x2_0<br/>PP-LCNet_x2_5<br/>PP-LCNetV2_small<br/>PP-LCNetV2_base<br/>PP-LCNetV2_large<br/>ResNet18<br/>ResNet18_vd<br/>ResNet34<br/>ResNet34_vd<br/>ResNet50<br/>ResNet50_vd<br/>ResNet101<br/>ResNet101_vd<br/>ResNet152<br/>ResNet152_vd<br/>ResNet200_vd<br/>SwinTransformer_tiny_patch4_window7_224<br/>SwinTransformer_small_patch4_window7_224<br/>SwinTransformer_base_patch4_window7_224<br/>SwinTransformer_base_patch4_window12_384<br/>SwinTransformer_large_patch4_window7_224<br/>SwinTransformer_large_patch4_window12_384<br/>StarNet-S1<br/>StarNet-S2<br/>StarNet-S3<br/>StarNet-S4<br/>FasterNet-L<br/>FasterNet-M<br/>FasterNet-S<br/>FasterNet-T0<br/>FasterNet-T1<br/>FasterNet-T2</details></td>
   <tr>
     <td>文档图像方向分类</td>
     <td>PP-LCNet_x1_0_doc_ori</td>

docs/tutorials/models/support_model_list.md

@@ -57,13 +57,21 @@
 | MobileNetV3_large_x0_75 | [MobileNetV3_large_x0_75.yaml](../../../paddlex/configs/image_classification/MobileNetV3_large_x0_75.yaml)|
 | MobileNetV3_large_x1_0 | [MobileNetV3_large_x1_0.yaml](../../../paddlex/configs/image_classification/MobileNetV3_large_x1_0.yaml)|
 | MobileNetV3_large_x1_25 | [MobileNetV3_large_x1_25.yaml](../../../paddlex/configs/image_classification/MobileNetV3_large_x1_25.yaml)|
-### 6.PP-HGNet 系列
+### 6.MobileNetV4 系列
+| 模型名称 | config |
+| :--- | :---: |
+| MobileNetV4_conv_small | [MobileNetV4_conv_small.yaml](../../../paddlex/configs/image_classification/MobileNetV4_conv_small.yaml)|
+| MobileNetV4_conv_medium | [MobileNetV4_conv_medium.yaml](../../../paddlex/configs/image_classification/MobileNetV4_conv_medium.yaml)|
+| MobileNetV4_conv_large | [MobileNetV4_conv_large.yaml](../../../paddlex/configs/image_classification/MobileNetV4_conv_large.yaml)|
+| MobileNetV4_hybrid_medium | [MobileNetV4_hybrid_medium.yaml](../../../paddlex/configs/image_classification/MobileNetV4_hybrid_medium.yaml)|
+| MobileNetV4_hybrid_large | [MobileNetV4_hybrid_large.yaml](../../../paddlex/configs/image_classification/MobileNetV4_hybrid_large.yaml)|
+### 7.PP-HGNet 系列
 | 模型名称 | config |
 | :--- | :---: |
 | PP-HGNet_tiny | [PP-HGNet_tiny.yaml](../../../paddlex/configs/image_classification/PP-HGNet_tiny.yaml)|
 | PP-HGNet_small | [PP-HGNet_small.yaml](../../../paddlex/configs/image_classification/PP-HGNet_small.yaml)|
 | PP-HGNet_base | [PP-HGNet_base.yaml](../../../paddlex/configs/image_classification/PP-HGNet_base.yaml)|
-### 7.PP-HGNetV2 系列
+### 8.PP-HGNetV2 系列
 | 模型名称 | config |
 | :--- | :---: |
 | PP-HGNetV2-B0 | [PP-HGNetV2-B0.yaml](../../../paddlex/configs/image_classification/PP-HGNetV2-B0.yaml)|
@@ -73,12 +81,12 @@
 | PP-HGNetV2-B4 | [PP-HGNetV2-B4.yaml](../../../paddlex/configs/image_classification/PP-HGNetV2-B4.yaml)|
 | PP-HGNetV2-B5 | [PP-HGNetV2-B5.yaml](../../../paddlex/configs/image_classification/PP-HGNetV2-B5.yaml)|
 | PP-HGNetV2-B6 | [PP-HGNetV2-B6.yaml](../../../paddlex/configs/image_classification/PP-HGNetV2-B6.yaml)|
-### 8.CLIP 系列
+### 9.CLIP 系列
 | 模型名称 | config |
 | :--- | :---: |
 | CLIP_vit_base_patch16_224 | [CLIP_vit_base_patch16_224.yaml](../../../paddlex/configs/image_classification/CLIP_vit_base_patch16_224.yaml)|
 | CLIP_vit_large_patch14_224 | [CLIP_vit_large_patch14_224.yaml](../../../paddlex/configs/image_classification/CLIP_vit_large_patch14_224.yaml)|
-### 9.ConvNeXt 系列
+### 10.ConvNeXt 系列
 | 模型名称 | config |
 | :--- | :---: |
 | ConvNeXt_tiny | [ConvNeXt_tiny.yaml](../../../paddlex/configs/image_classification/ConvNeXt_tiny.yaml)|
@@ -87,7 +95,7 @@
 | ConvNeXt_base_384 | [ConvNeXt_base_384.yaml](../../../paddlex/configs/image_classification/ConvNeXt_base_384.yaml)|
 | ConvNeXt_large_224 | [ConvNeXt_large_224.yaml](../../../paddlex/configs/image_classification/ConvNeXt_large_224.yaml)|
 | ConvNeXt_large_384 | [ConvNeXt_large_384.yaml](../../../paddlex/configs/image_classification/ConvNeXt_large_384.yaml)|
-### 10.SwinTransformer系列
+### 11.SwinTransformer系列
 | 模型名称 | config |
 | :--- | :---: |
 | SwinTransformer_tiny_patch4_window7_224 | [SwinTransformer_tiny_patch4_window7_224.yaml](../../../paddlex/configs/image_classification/SwinTransformer_tiny_patch4_window7_224.yaml)|
@@ -96,8 +104,24 @@
 | SwinTransformer_base_patch4_window12_384 | [SwinTransformer_base_patch4_window12_384.yaml](../../../paddlex/configs/image_classification/SwinTransformer_base_patch4_window12_384.yaml)|
 | SwinTransformer_large_patch4_window7_224 | [SwinTransformer_large_patch4_window7_224.yaml](../../../paddlex/configs/image_classification/SwinTransformer_large_patch4_window7_224.yaml)|
 | SwinTransformer_large_patch4_window12_384 | [SwinTransformer_large_patch4_window12_384.yaml](../../../paddlex/configs/image_classification/SwinTransformer_large_patch4_window12_384.yaml)|
+### 12.StarNet系列
+| 模型名称 | config |
+| :--- | :---: |
+| StarNet-S1 | [StarNet-S1.yaml](../../../paddlex/configs/image_classification/StarNet-S1.yaml)|
+| StarNet-S2 | [StarNet-S2.yaml](../../../paddlex/configs/image_classification/StarNet-S2.yaml)|
+| StarNet-S3 | [StarNet-S3.yaml](../../../paddlex/configs/image_classification/StarNet-S3.yaml)|
+| StarNet-S4 | [StarNet-S4.yaml](../../../paddlex/configs/image_classification/StarNet-S4.yaml)|
+### 13.FasterNet系列
+| 模型名称 | config |
+| :--- | :---: |
+| FasterNet-L | [FasterNet-L.yaml](../../../paddlex/configs/image_classification/FasterNet-L.yaml)|
+| FasterNet-M | [FasterNet-M.yaml](../../../paddlex/configs/image_classification/FasterNet-M.yaml)|
+| FasterNet-S | [FasterNet-S.yaml](../../../paddlex/configs/image_classification/FasterNet-S.yaml)|
+| FasterNet-T0 | [FasterNet-T0.yaml](../../../paddlex/configs/image_classification/FasterNet-T0.yaml)|
+| FasterNet-T1 | [FasterNet-T1.yaml](../../../paddlex/configs/image_classification/FasterNet-T1.yaml)|
+| FasterNet-T2 | [FasterNet-T2.yaml](../../../paddlex/configs/image_classification/FasterNet-T2.yaml)|
 
-### 11. 多标签分类系列
+### 14. 多标签分类系列
 | 模型名称 | config |
 | :--- | :---: |
 | ResNet50_ML | [ResNet50_ML.yaml](../../../paddlex/configs/multilabel_classification/ResNet50_ML.yaml)|

paddlex/configs/image_classification/FasterNet-L.yaml

@@ -0,0 +1,40 @@
+Global:
+  model: FasterNet-L
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 64
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/FasterNet_L_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/FasterNet-M.yaml

@@ -0,0 +1,40 @@
+Global:
+  model: FasterNet-M
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 64
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/FasterNet_M_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/FasterNet-S.yaml

@@ -0,0 +1,40 @@
+Global:
+  model: FasterNet-S
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/FasterNet_S_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/FasterNet-T0.yaml

@@ -0,0 +1,40 @@
+Global:
+  model: FasterNet-T0
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/FasterNet_T0_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/FasterNet-T1.yaml

@@ -0,0 +1,40 @@
+Global:
+  model: FasterNet-T1
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/FasterNet_T1_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/FasterNet-T2.yaml

@@ -0,0 +1,40 @@
+Global:
+  model: FasterNet-T2
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/FasterNet_T2_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/MobileNetV4_conv_large.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: MobileNetV4_conv_large
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 64
+  learning_rate: 0.13
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/MobileNetV4_conv_large_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/MobileNetV4_conv_medium.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: MobileNetV4_conv_medium
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 0.13
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/MobileNetV4_conv_medium_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/MobileNetV4_conv_small.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: MobileNetV4_conv_small
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 0.13
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/MobileNetV4_conv_small_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/MobileNetV4_hybrid_large.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: MobileNetV4_hybrid_large
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 64
+  learning_rate: 0.13
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/MobileNetV4_hybrid_large_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/MobileNetV4_hybrid_medium.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: MobileNetV4_hybrid_medium
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 64
+  learning_rate: 0.13
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/MobileNetV4_hybrid_medium_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/StarNet-S1.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: StarNet-S1
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/StarNet_S1_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/StarNet-S2.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: StarNet-S2
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/StarNet_S2_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/StarNet-S3.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: StarNet-S3
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/StarNet_S3_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/configs/image_classification/StarNet-S4.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: StarNet-S4
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"
+  device: gpu:0,1,2,3
+  output: "output"
+
+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_type: null
+  split: 
+    enable: False
+    train_percent: null
+    val_percent: null
+
+Train:
+  num_classes: 102
+  epochs_iters: 20
+  batch_size: 128
+  learning_rate: 3e-3
+  pretrain_weight_path: null
+  warmup_steps: 5
+  resume_path: null
+  log_interval: 1
+  eval_interval: 1
+  save_interval: 1
+
+Evaluate:
+  weight_path: "output/best_model.pdparams"
+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/StarNet_S4_pretrained.pdparams
+
+Predict:
+  model_dir: "output/best_model"
+  input_path: "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg"
+  kernel_option:
+    run_mode: paddle
+    batch_size: 1

paddlex/modules/base/predictor/utils/official_models.py

@@ -103,6 +103,21 @@ SwinTransformer_large_patch4_window12_384_infer.tar",
     "PP-HGNetV2-B4": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/PP-HGNetV2-B4_infer.tar",
     "PP-HGNetV2-B5": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/PP-HGNetV2-B5_infer.tar",
     "PP-HGNetV2-B6": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/PP-HGNetV2-B6_infer.tar",
+    "FasterNet-L": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/FasterNet-L_infer.tar",
+    "FasterNet-M": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/FasterNet-M_infer.tar",
+    "FasterNet-S": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/FasterNet-S_infer.tar",
+    "FasterNet-T0": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/FasterNet-T0_infer.tar",
+    "FasterNet-T1": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/FasterNet-T1_infer.tar",
+    "FasterNet-T2": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/FasterNet-T2_infer.tar",
+    "StarNet-S1": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/StarNet-S1_infer.tar",
+    "StarNet-S2": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/StarNet-S2_infer.tar",
+    "StarNet-S3": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/StarNet-S3_infer.tar",
+    "StarNet-S4": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/StarNet-S4_infer.tar",
+    "MobileNetV4_conv_small": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/MobileNetV4_conv_small_infer.tar",
+    "MobileNetV4_conv_medium": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/MobileNetV4_conv_medium_infer.tar",
+    "MobileNetV4_conv_large": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/MobileNetV4_conv_large_infer.tar",
+    "MobileNetV4_hybrid_medium": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/MobileNetV4_hybrid_medium_infer.tar",
+    "MobileNetV4_hybrid_large": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/MobileNetV4_hybrid_large_infer.tar",
     "CLIP_vit_base_patch16_224": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/\
 CLIP_vit_base_patch16_224_infer.tar",
     "CLIP_vit_large_patch14_224": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1/\

paddlex/modules/image_classification/model_list.py

@@ -40,6 +40,11 @@ MODELS = [
     "MobileNetV3_small_x0_75",
     "MobileNetV3_small_x1_0",
     "MobileNetV3_small_x1_25",
+    "MobileNetV4_conv_small",
+    "MobileNetV4_conv_medium",
+    "MobileNetV4_conv_large",
+    "MobileNetV4_hybrid_medium",
+    "MobileNetV4_hybrid_large",
     "PP-HGNet_tiny",
     "PP-HGNet_small",
     "PP-HGNet_base",
@@ -79,4 +84,14 @@ MODELS = [
     "SwinTransformer_base_patch4_window12_384",
     "SwinTransformer_large_patch4_window7_224",
     "SwinTransformer_large_patch4_window12_384",
+    "StarNet-S1",
+    "StarNet-S2",
+    "StarNet-S3",
+    "StarNet-S4",
+    "FasterNet-L",
+    "FasterNet-M",
+    "FasterNet-S",
+    "FasterNet-T0",
+    "FasterNet-T1",
+    "FasterNet-T2",
 ]

paddlex/modules/semantic_segmentation/dataset_checker/__init__.py

@@ -38,7 +38,7 @@ class SegDatasetChecker(BaseDatasetChecker):
             str: the root directory of converted dataset.
         """
         return convert_dataset(
-            self.check_dataset_config.src_dataset_type, src_dataset_dir
+            self.check_dataset_config.convert.src_dataset_type, src_dataset_dir
         )
 
     def split_dataset(self, src_dataset_dir: str) -> str:

paddlex/repo_apis/PaddleClas_api/cls/model.py

@@ -121,7 +121,6 @@ class ClsModel(BaseModel):
 
             config.dump(config_path)
             self._assert_empty_kwargs(kwargs)
-
             return self.runner.train(
                 config_path, cli_args, device, ips, save_dir, do_eval=do_eval
             )

paddlex/repo_apis/PaddleClas_api/cls/register.py

@@ -664,6 +664,51 @@ register_model_info(
 
 register_model_info(
     {
+        "model_name": "MobileNetV4_conv_small",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "MobileNetV4_conv_small.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "MobileNetV4_conv_medium",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "MobileNetV4_conv_medium.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "MobileNetV4_conv_large",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "MobileNetV4_conv_large.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "MobileNetV4_hybrid_medium",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "MobileNetV4_hybrid_medium.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "MobileNetV4_hybrid_large",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "MobileNetV4_hybrid_large.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
         "model_name": "ConvNeXt_tiny",
         "suite": "Cls",
         "config_path": osp.join(PDX_CONFIG_DIR, "ConvNeXt_tiny.yaml"),
@@ -724,6 +769,96 @@ register_model_info(
 
 register_model_info(
     {
+        "model_name": "StarNet-S1",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "StarNet-S1.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "StarNet-S2",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "StarNet-S2.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "StarNet-S3",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "StarNet-S3.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "StarNet-S4",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "StarNet-S4.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "FasterNet-L",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "FasterNet-L.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "FasterNet-M",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "FasterNet-M.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "FasterNet-S",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "FasterNet-S.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "FasterNet-T0",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "FasterNet-T0.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "FasterNet-T1",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "FasterNet-T1.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
+        "model_name": "FasterNet-T2",
+        "suite": "Cls",
+        "config_path": osp.join(PDX_CONFIG_DIR, "FasterNet-T2.yaml"),
+        "supported_apis": ["train", "evaluate", "predict", "export"],
+    }
+)
+
+register_model_info(
+    {
         "model_name": "PP-LCNet_x1_0_ML",
         "suite": "Cls",
         "config_path": osp.join(PDX_CONFIG_DIR, "PP-LCNet_x1_0_ML.yaml"),

paddlex/repo_apis/PaddleClas_api/configs/FasterNet-L.yaml

@@ -0,0 +1,163 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: False
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: FasterNet_L
+  class_num: 1000
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.05
+  clip_grad: 0.01
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0006
+    warmup_start_lr: 0.000001
+    warmup_epoch: 20
+    eta_min: 0.00001
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m7-mstd0.5-inc1
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.7
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            to_np: False
+            channel_first: False
+        - ResizeImage:  
+            interpolation: bicubic
+            backend: pil
+            resize_short: 248
+        - CropImage: 
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 4
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/FasterNet-M.yaml

@@ -0,0 +1,163 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: False
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: FasterNet_M
+  class_num: 1000
+  
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.05
+  clip_grad: 1
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0006 
+    warmup_start_lr: 0.000001
+    warmup_epoch: 20
+    eta_min: 0.00001
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m7-mstd0.5-inc1 
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.5
+              prob: 0.5
+            CutmixOperator:
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            to_np: False
+            channel_first: False
+        - ResizeImage:  
+            interpolation: bicubic
+            backend: pil
+            resize_short: 248
+        - CropImage: 
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 4
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/FasterNet-S.yaml

@@ -0,0 +1,163 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: False
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: FasterNet_S
+  class_num: 1000
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.03
+  clip_grad: null
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0006
+    warmup_start_lr: 0.000001
+    warmup_epoch: 20
+    eta_min: 0.00001
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m7-mstd0.5-inc1
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.3
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            to_np: False
+            channel_first: False
+        - ResizeImage:  ##RandomResized
+            interpolation: bicubic
+            backend: pil
+            resize_short: 248
+        - CropImage: 
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 6
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/FasterNet-T0.yaml

@@ -0,0 +1,161 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: False
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: FasterNet_T0
+  class_num: 1000
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.005
+  clip_grad: null
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0006 # 8gpu
+    warmup_start_lr: 0.000001
+    warmup_epoch: 20
+    eta_min: 0.00001
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.05
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            to_np: False
+            channel_first: False
+        - ResizeImage: 
+            interpolation: bicubic
+            backend: pil
+            resize_short: 248
+        - CropImage: 
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 6
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/FasterNet-T1.yaml

@@ -0,0 +1,163 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: False
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: FasterNet_T1
+  class_num: 1000
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.01
+  clip_grad: null
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0006
+    warmup_start_lr: 0.000001
+    warmup_epoch: 20
+    eta_min: 0.00001
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m3-mstd0.5-inc1 
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.05
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            to_np: False
+            channel_first: False
+        - ResizeImage:  
+            interpolation: bicubic
+            backend: pil
+            resize_short: 248
+        - CropImage:
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 6
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/FasterNet-T2.yaml

@@ -0,0 +1,162 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: False
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: FasterNet_T2
+  class_num: 1000
+
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.02
+  clip_grad: null
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0006
+    warmup_start_lr: 0.000001
+    warmup_epoch: 20
+    eta_min: 0.00001
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m5-mstd0.5-inc1 
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.1
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64 
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            to_np: False
+            channel_first: False
+        - ResizeImage:  
+            interpolation: bicubic
+            backend: pil
+            resize_short: 248
+        - CropImage: 
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 6
+      use_shared_memory: True
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/MobileNetV4_conv_large.yaml

@@ -0,0 +1,181 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 600
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 448, 448]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: MobileNetV4_conv_large
+  drop_rate: 0.2
+  drop_path_rate: 0.35
+  class_num: 1000
+ 
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.2
+  clip_grad: 5.0
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    # for 8 cards
+    name: Cosine
+    learning_rate: 0.002 ##null
+    eta_min: 1.0e-06
+    warmup_epoch: 20
+    warmup_start_lr: 0
+
+
+EMA:
+  decay: 0.9998 
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 384
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m8-inc1-mstd1.0
+            interpolation: bicubic
+            img_size: 256
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            sl: 0.02
+            sh: 1.0/3.0
+            r1: 0.3
+            attempt: 10
+            use_log_aspect: True
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: # todo 确认是否需要
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 256
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - ResizeImage:
+            backend: pil
+            interpolation: bicubic
+            resize_short: 448
+        - CropImage:
+            size: 448
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/MobileNetV4_conv_medium.yaml

@@ -0,0 +1,181 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 500
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 320, 320]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: MobileNetV4_conv_medium
+  drop_rate: 0.2
+  drop_path_rate: 0.1
+  class_num: 1000
+ 
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.1
+  clip_grad: 5.0
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    # for 8 cards
+    name: Cosine
+    learning_rate: 0.002
+    eta_min: 0
+    warmup_epoch: 20
+    warmup_start_lr: 0
+
+
+EMA:
+  decay: 0.9998 
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 256
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m8-inc1-mstd1.0
+            interpolation: bicubic
+            img_size: 256
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            sl: 0.02
+            sh: 1.0/3.0
+            r1: 0.3
+            attempt: 10
+            use_log_aspect: True
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 512
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - ResizeImage:
+            backend: pil
+            interpolation: bicubic
+            resize_short: 320
+        - CropImage:
+            size: 320
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/MobileNetV4_conv_small.yaml

@@ -0,0 +1,181 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 2400
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 256, 256]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: MobileNetV4_conv_small
+  drop_rate: 0.25
+  drop_path_rate: 0.03
+  class_num: 1000
+ 
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.06
+  clip_grad: 5.0
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    # for 8 cards
+    name: Cosine
+    learning_rate: 0.002 ##null
+    eta_min: 0
+    warmup_epoch: 5
+    warmup_start_lr: 0
+
+
+EMA:
+  decay: 0.9998 
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m8-inc1-mstd1.0
+            interpolation: bicubic
+            img_size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            sl: 0.02
+            sh: 1.0/3.0
+            r1: 0.3
+            attempt: 10
+            use_log_aspect: True
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 1024
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - ResizeImage:
+            backend: pil
+            interpolation: bicubic
+            resize_short: 256
+        - CropImage:
+            size: 256
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/MobileNetV4_hybrid_large.yaml

@@ -0,0 +1,181 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 600
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 448, 448]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: True
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: MobileNetV4_hybrid_large
+  drop_rate: 0.2
+  drop_path_rate: 0.1
+  class_num: 1000
+ 
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.1
+  clip_grad: 5.0
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    # for 8 cards
+    name: Cosine
+    learning_rate: 0.002
+    eta_min: 1.0e-06
+    warmup_epoch: 20
+    warmup_start_lr: 0
+
+
+EMA:
+  decay: 0.9998 
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 384
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m9-inc1-mstd1.0
+            interpolation: bicubic
+            img_size: 384
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            sl: 0.02
+            sh: 1.0/3.0
+            r1: 0.3
+            attempt: 10
+            use_log_aspect: True
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 192
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list_test.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - ResizeImage:
+            backend: pil
+            interpolation: bicubic
+            resize_short: 448
+        - CropImage:
+            size: 448
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 448
+    - CropImage:
+        size: 448
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/MobileNetV4_hybrid_medium.yaml

@@ -0,0 +1,175 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 500
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 256, 256]
+  save_inference_dir: ./inference
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: True
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: MobileNetV4_hybrid_medium
+  drop_rate: 0.2
+  drop_path_rate: 0.1
+  class_num: 1000
+ 
+ 
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.1
+  clip_grad: 5.0
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    # for 8 cards
+    name: Cosine
+    learning_rate: 0.001
+    eta_min: 0
+    warmup_epoch: 20
+    warmup_start_lr: 0
+
+EMA:
+  decay: 0.9998 
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 224
+            interpolation: bicubic
+            backend: pil
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m9-inc1-mstd1.0
+            interpolation: bicubic
+            img_size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            sl: 0.02
+            sh: 1.0/3.0
+            r1: 0.3
+            attempt: 10
+            use_log_aspect: True
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: # todo 确认是否需要
+              alpha: 1.0
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 512
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list_test.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - ResizeImage:
+            backend: pil
+            interpolation: bicubic
+            resize_short: 256
+        - CropImage:
+            size: 256
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 64
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 12
+      use_shared_memory: True
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 256
+    - CropImage:
+        size: 256
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/StarNet-S1.yaml

@@ -0,0 +1,165 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: StarNet_S1
+  drop_rate: 0
+  drop_path_rate: 0
+  class_num: 1000
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.05
+  clip_grad: None
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 3e-3
+    eta_min: 1e-5
+    warmup_epoch: 5
+    warmup_start_lr: 1e-6
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 224
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m1-mstd0.5-inc1
+            interpolation: random
+            img_size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 0.2
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 256
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - ResizeImage:
+            interpolation: bicubic
+            backend: pil
+            resize_short: 256
+        - CropImage:
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 20
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 224
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/StarNet-S2.yaml

@@ -0,0 +1,165 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: StarNet_S2
+  drop_rate: 0
+  drop_path_rate: 0.01
+  class_num: 1000
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.05
+  clip_grad: None
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 3e-3
+    eta_min: 1e-5
+    warmup_epoch: 5
+    warmup_start_lr: 1e-6
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 224
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m1-mstd0.5-inc1
+            interpolation: random
+            img_size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 0.2
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 256
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - ResizeImage:
+            interpolation: bicubic
+            backend: pil
+            resize_short: 256
+        - CropImage:
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 224
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/StarNet-S3.yaml

@@ -0,0 +1,165 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: StarNet_S3
+  drop_rate: 0
+  drop_path_rate: 0.01
+  class_num: 1000
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.05
+  clip_grad: None
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 3e-3
+    eta_min: 1e-5
+    warmup_epoch: 5
+    warmup_start_lr: 1e-6
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list_smallbatch.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 224
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m1-mstd0.5-inc1
+            interpolation: random
+            img_size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator:
+              alpha: 0.2
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 256
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - ResizeImage:
+            interpolation: bicubic
+            backend: pil
+            resize_short: 256
+        - CropImage:
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 224
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]

paddlex/repo_apis/PaddleClas_api/configs/StarNet-S4.yaml

@@ -0,0 +1,165 @@
+# global configs
+Global:
+  checkpoints: null
+  pretrained_model: null
+  output_dir: ./output/
+  device: gpu
+  save_interval: 1
+  eval_during_train: True
+  eval_interval: 1
+  epochs: 300
+  print_batch_step: 10
+  use_visualdl: False
+  # used for static mode and model export
+  image_shape: [3, 224, 224]
+  save_inference_dir: ./inference
+
+
+# mixed precision
+AMP:
+  use_amp: True
+  use_fp16_test: False
+  scale_loss: 128.0
+  use_dynamic_loss_scaling: True
+  use_promote: False
+  # O1: mixed fp16, O2: pure fp16
+  level: O1
+
+
+# model architecture
+Arch:
+  name: StarNet_S4
+  drop_rate: 0
+  drop_path_rate: 0.02
+  class_num: 1000
+
+# loss function config for traing/eval process
+Loss:
+  Train:
+    - CELoss:
+        weight: 1.0
+        epsilon: 0.1
+  Eval:
+    - CELoss:
+        weight: 1.0
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1e-8
+  weight_decay: 0.05
+  clip_grad: None
+  no_weight_decay_name: null
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 3e-3
+    eta_min: 1e-5
+    warmup_epoch: 5
+    warmup_start_lr: 1e-6
+
+
+# data loader for train and eval
+DataLoader:
+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/train_list.txt
+      transform_ops:
+        - DecodeImage:
+            to_rgb: True
+            channel_first: False
+        - RandCropImage:
+            size: 224
+        - RandFlipImage:
+            flip_code: 1
+        - TimmAutoAugment:
+            config_str: rand-m1-mstd0.5-inc1
+            interpolation: random
+            img_size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: ''
+        - RandomErasing:
+            EPSILON: 0.25
+            mode: pixel
+      batch_transform_ops:
+        - OpSampler:
+            MixupOperator:
+              alpha: 0.8
+              prob: 0.5
+            CutmixOperator: 
+              alpha: 0.2
+              prob: 0.5
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 256
+      drop_last: False
+      shuffle: True
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+  Eval:
+    dataset: 
+      name: ImageNetDataset
+      image_root: ./dataset/ILSVRC2012/
+      cls_label_path: ./dataset/ILSVRC2012/val_list.txt
+      transform_ops:
+        - DecodeImage:
+            backend: pil
+            channel_first: False
+        - ResizeImage:
+            interpolation: bicubic
+            backend: pil
+            resize_short: 256
+        - CropImage:
+            size: 224
+        - NormalizeImage:
+            scale: 1.0/255.0
+            mean: [0.485, 0.456, 0.406]
+            std: [0.229, 0.224, 0.225]
+            order: 'hwc'
+    sampler:
+      name: DistributedBatchSampler
+      batch_size: 4
+      drop_last: False
+      shuffle: False
+    loader:
+      num_workers: 4
+      use_shared_memory: False
+
+Infer:
+  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
+  batch_size: 10
+  transforms:
+    - DecodeImage:
+        to_rgb: True
+        channel_first: False
+    - ResizeImage:
+        resize_short: 224
+    - CropImage:
+        size: 224
+    - NormalizeImage:
+        scale: 1.0/255.0
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+        order: ''
+    - ToCHWImage:
+  PostProcess:
+    name: Topk
+    topk: 5
+    class_id_map_file: ppcls/utils/imagenet1k_label_list.txt
+
+Metric:
+  Train:
+    - TopkAcc:
+        topk: [1, 5]
+  Eval:
+    - TopkAcc:
+        topk: [1, 5]