|
|
@@ -59,11 +59,20 @@ class WiredTableVisualizer:
|
|
|
Args:
|
|
|
hpred_up: 横线预测mask（上采样后）
|
|
|
vpred_up: 竖线预测mask（上采样后）
|
|
|
- bboxes: 单元格bbox列表
|
|
|
+ bboxes: 单元格bbox列表（原图坐标）
|
|
|
upscale: 上采样比例
|
|
|
output_path: 输出路径
|
|
|
"""
|
|
|
h, w = hpred_up.shape[:2]
|
|
|
+
|
|
|
+ # 调试：验证上采样图像尺寸
|
|
|
+ expected_h = int(bboxes[-1][3] * upscale + 0.5) if bboxes else 0
|
|
|
+ expected_w = int(bboxes[-1][2] * upscale + 0.5) if bboxes else 0
|
|
|
+ logger.debug(
|
|
|
+ f"上采样图像尺寸: 实际=[{h}, {w}], "
|
|
|
+ f"预期(基于最大bbox)≈[{expected_h}, {expected_w}], "
|
|
|
+ f"upscale={upscale:.3f}"
|
|
|
+ )
|
|
|
|
|
|
# 与连通域提取相同的预处理，以获得直观的网格线背景
|
|
|
_, h_bin = cv2.threshold(hpred_up, 127, 255, cv2.THRESH_BINARY)
|
|
|
@@ -78,12 +87,32 @@ class WiredTableVisualizer:
|
|
|
vis[grid_mask > 0] = [0, 0, 255] # 红色线条
|
|
|
|
|
|
# 在上采样坐标系上绘制单元格框
|
|
|
- for box in bboxes:
|
|
|
- x1, y1, x2, y2 = [int(c * upscale) for c in box]
|
|
|
- cv2.rectangle(vis, (x1, y1), (x2, y2), (0, 255, 0), 2)
|
|
|
+ # 修复：使用更精确的坐标转换，避免累积误差
|
|
|
+ for idx, box in enumerate(bboxes):
|
|
|
+ # 使用四舍五入而不是直接截断，提高精度
|
|
|
+ x1 = int(box[0] * upscale + 0.5)
|
|
|
+ y1 = int(box[1] * upscale + 0.5)
|
|
|
+ x2 = int(box[2] * upscale + 0.5)
|
|
|
+ y2 = int(box[3] * upscale + 0.5)
|
|
|
+
|
|
|
+ # 确保坐标在图像范围内
|
|
|
+ x1 = max(0, min(x1, w - 1))
|
|
|
+ y1 = max(0, min(y1, h - 1))
|
|
|
+ x2 = max(0, min(x2, w - 1))
|
|
|
+ y2 = max(0, min(y2, h - 1))
|
|
|
+
|
|
|
+ if x2 > x1 and y2 > y1:
|
|
|
+ cv2.rectangle(vis, (x1, y1), (x2, y2), (0, 255, 0), 2)
|
|
|
+
|
|
|
+ # 调试日志：输出前几个和后几个单元格的坐标转换信息
|
|
|
+ if idx < 3 or idx >= len(bboxes) - 3:
|
|
|
+ logger.debug(
|
|
|
+ f"单元格 {idx}: 原图坐标 [{box[0]:.1f}, {box[1]:.1f}, {box[2]:.1f}, {box[3]:.1f}] "
|
|
|
+ f"-> 上采样坐标 [{x1}, {y1}, {x2}, {y2}] (upscale={upscale:.3f})"
|
|
|
+ )
|
|
|
|
|
|
cv2.imwrite(output_path, vis)
|
|
|
- logger.info(f"连通域可视化: {output_path}")
|
|
|
+ logger.info(f"连通域可视化: {output_path} (共 {len(bboxes)} 个单元格)")
|
|
|
|
|
|
@staticmethod
|
|
|
def visualize_grid_structure(
|
