YongZ-Lee

遇到训练能跑几十个epoch后，突然在test module时候会报在/utils/ocr_metric/icdar2015/detection/iou.py的第12行的“rotatedRectangleIntersection(rect_a, rect_b)”错误，找不到错误的头绪

找到/post_processing/seg_detector_representer.py，找到def boxes_from_bitmap(),在105行， 原本的“contours, _ = cv2.findContours((bitmap * 255).astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)”改成， “_, contours, _ = cv2.findContours((bitmap * 255).astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)”

数据集和关于graph_edge

2

您好，想问问提供的数据集标签里面的test.txt, train.txt, val.txt都有提到graph_edge，但是这再哪里呢

Failed to download the provided pre-training weights

5

Thank you for your work, I found that the provided pre-training downloads failed, such as 'yolov8/runs/detect/yolov8s-custom-detection/weights/' and 'yolov8/runs/detect/yolov8s-custom-structure/weights/' The weights are failing, can you check this, thank you so much!

The structure recognition of UniTable that I trained is supposed to output <td> [] </td> after inference, but it only shows <td> </td>

3

Thank you very much for your work. I trained both the base and large versions of UniTable using the complete PubTabNet dataset, and I wanted to test my own model...