training is stuck on a fixed value

[madinwei]

@yanx27 , Hello yan, I am facing a problem when I train the model the Instance accuracy and class accuracy are stuck on a fixed value. Epoch 4 (17/200): 100%|█████████████████████████████████████████| 410/410 [01:18<00:00, 5.24it/s] Train Instance Accuracy: 0.063618 100%|█████████████████████████████████████████| 103/103 [00:14<00:00, 7.28it/s] Test Instance Accuracy: 0.040453, Class Accuracy: 0.025000 Best Instance Accuracy: 0.040453, Class Accuracy: 0.025000 Saving at log/classification/pointnet2_cls_ssg/checkpoints/best_model.pth Epoch 5 (18/200): 100%|█████████████████████████████████████████| 410/410 [01:16<00:00, 5.38it/s] Train Instance Accuracy: 0.063516 100%|█████████████████████████████████████████| 103/103 [00:14<00:00, 7.30it/s] Test Instance Accuracy: 0.040453, Class Accuracy: 0.025000 Best Instance Accuracy: 0.040453, Class Accuracy: 0.025000 Saving at log/classification/pointnet2_cls_ssg/checkpoints/best_model.pth no matter how much I keep it train. I tried both in windows and ubuntu. I used both the latest version of PyTorch=1.12 and 1.6 I am using RTX3060 . I have no idea why the training is not changing.

I hope anyone can provide any help to this problem, and thank you in advance

[xiximelon]

Hello,have you found the solution ? the same problem met by me.

[yzlf625]

我也遇到了相同的问题，直至训练结束全程保持了几乎相同的准确率，请问现在解决了嘛