训练过程只保存pth文件，并没有真正的训练

[liliwannian]

您好，感谢您的贡献。我用自己的数据集在复现您的代码的时候，出现以下问题。一共有六千多张图，训练阶段，只保存pth文件，并没有进行真正的训练，几分钟就完成了训练。 2021-09-23 10:51:02,950 - mmdet - INFO - workflow: [('train', 1)], max: 20 epochs 2021-09-23 10:51:18,827 - mmdet - INFO - Saving checkpoint at 1 epochs 2021-09-23 10:51:34,376 - mmdet - INFO - Saving checkpoint at 2 epochs 2021-09-23 10:51:49,927 - mmdet - INFO - Saving checkpoint at 3 epochs 2021-09-23 10:52:05,186 - mmdet - INFO - Saving checkpoint at 4 epochs 2021-09-23 10:52:20,482 - mmdet - INFO - Saving checkpoint at 5 epochs 2021-09-23 10:52:35,692 - mmdet - INFO - Saving checkpoint at 6 epochs 2021-09-23 10:52:50,909 - mmdet - INFO - Saving checkpoint at 7 epochs 2021-09-23 10:53:06,247 - mmdet - INFO - Saving checkpoint at 8 epochs 2021-09-23 10:53:21,548 - mmdet - INFO - Saving checkpoint at 9 epochs 2021-09-23 10:53:36,903 - mmdet - INFO - Saving checkpoint at 10 epochs 2021-09-23 10:53:52,295 - mmdet - INFO - Saving checkpoint at 11 epochs 2021-09-23 10:54:07,654 - mmdet - INFO - Saving checkpoint at 12 epochs 2021-09-23 10:54:23,017 - mmdet - INFO - Saving checkpoint at 13 epochs 2021-09-23 10:54:38,409 - mmdet - INFO - Saving checkpoint at 14 epochs 2021-09-23 10:54:53,867 - mmdet - INFO - Saving checkpoint at 15 epochs 2021-09-23 10:55:09,329 - mmdet - INFO - Saving checkpoint at 16 epochs 2021-09-23 10:55:24,748 - mmdet - INFO - Saving checkpoint at 17 epochs 2021-09-23 10:55:40,221 - mmdet - INFO - Saving checkpoint at 18 epochs 2021-09-23 10:55:55,771 - mmdet - INFO - Saving checkpoint at 19 epochs 2021-09-23 10:56:11,343 - mmdet - INFO - Saving checkpoint at 20 epochs 我检查了切割图片生成的pkl文件，并没有问题。 希望您有空能回答，十分感谢！ #

[liliwannian]

https://github.com/jbwang1997/OBBDetection/issues/10#issue-980862061 类似于这个

[jbwang1997]

很有可能还是数据加载有问题，可以检查一下patch_annfile.pkl中的ann的bboxes是否有数据。 之后可以检查一下len(dataset)是否正常，在dataset中会过滤掉没有目标的图像块，如果len(dataset)为0，所有图像块都没有目标

[liliwannian]

非常感谢您的回复，问题已解决。现在我想把所有图片放进去训练，就是有目标和没目标都进行训练，请问应该怎么修改？希望您有空能够解答，感谢！