无法正确加载您给出的训练结果 about hanet-cd

[ChengxiHAN]

确实修改过哈，我想想看怎么解决。

[ChengxiHAN]

你也可以先训练几个epoch，看看能否测试。

[money6651626]

如果您方便的话可以直接提供HAN的训练权重pth，因为我看在open-cd的复现中直接引用了该类，这应该是您的最终版本），如果这是您论文上指标用的模型。（我相信在您的设备上应该是能够正确测试的，您通过torch.save(model.state_dict())的方式来重新保存权重，我通过torch.load_state_dict(torch.load("pth"))的方式直接按照键值对加载权重，这样不会受目录结构影响）

[work1-shekhaman]

@ChengxiHAN, did you found out, how to solve it, i am still facing this issue

[ChengxiHAN]

@ChengxiHAN, did you found out, how to solve it, i am still facing this issue

sorry, bro, it needs you to train HANet again. I haven't solved the problem yet.

[work1-shekhaman]

@ChengxiHAN, did you found out, how to solve it, i am still facing this issue

sorry, bro, it needs you to train HANet again. I haven't solved the problem yet.

If I train the model with batch size 4, predictions are in shape of (4,2,256,256) (for image size of (256,256))

Shape of tensor of predictions: torch.Size([4, 2, 256, 256])
Shape of tensor of target: torch.Size([1, 256, 256])
epoch 0 info 0 - 4: 0%| | 0/1245 [00:01<?, ?it/s]
Traceback (most recent call last):
File "trainHCX.py", line 157, in
cd_loss = criterion(cd_preds, labels)
File "/root/anaconda3/envs/hanet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/mnt/c/Users/91635/Desktop/Elm/gitrepo/HANet-CD-main/HANet-CD-main/utils/losses.py", line 96, in forward
bce = self.focal(predictions[0], target)
File "/root/anaconda3/envs/hanet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/mnt/c/Users/91635/Desktop/Elm/gitrepo/HANet-CD-main/HANet-CD-main/utils/losses.py", line 40, in forward
logpt = logpt.gather(1, target)
RuntimeError: Size does not match at dimension 0 expected index [262144, 1] to be smaller than self [512, 256] apart from dimension 1