twtygqyy / pytorch-srdensenet Goto Github PK

Epoch1: Loss: 116810211328.0000000000=>is it an error?

I just use dataset Yang 91 as the training dataset.

Does "30,000 images" in the result section include the images generated by data augmentation or not?

It seems if I include 30,000 images in the image folder, the script generate_train_srdensenet will run out of memory...

Thanks

I've trained model on my own datasets, the eval can only give me the result of PSNR and SSIM, how can i get the super-resolved for my other usage? Thank u

@twtygqyy I have run into trouble while I try to test the code. The error information is listed below. I have checked it but I can't correct it. Please give me some advise, thanks a lot.

D:\ProgramData\Anaconda3\python.exe D:/ProgramData/twt-pytorch-SRDenseNet-master/eval.py

Traceback (most recent call last):
File "D:/ProgramData/twt-pytorch-SRDenseNet-master/eval.py", line 31, in
model = torch.load(opt.model)["model"]
File "D:\ProgramData\Anaconda3\lib\site-packages\torch\serialization.py", line 303, in load
return _load(f, map_location, pickle_module)
File "D:\ProgramData\Anaconda3\lib\site-packages\torch\serialization.py", line 469, in _load
result = unpickler.load()
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 918: ordinal not in range(128)

I use your code, but replaced the dataset, but the loss was volatile and does not converge

I used https://github.com/twtygqyy/pytorch-SRResNet/tree/master/data in the README to generate the dataset.

but when I used the generated dataset and the model in the srdensenet.py to run.
python main.py --cuda
I got the RuntimeError: CUDNN_STATUS_BAD_PARAM
It seems that the sizeof input tensor do not match the model

Traceback (most recent call last):
File "main.py", line 161, in
main()
File "main.py", line 94, in main
train(training_data_loader, optimizer, model, criterion, epoch)
File "main.py", line 135, in train
out = model(input)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.py", line 206, in call
result = self.forward(*input, **kwargs)
File "/home/pzz/workspace/pytorch-SRDenseNet/srdensenet.py", line 101, in forward
residual = self.relu(self.lowlevel(x)) ==========> the input tensor is ((32L, 3L, 24L, 24L), while the
==========> required input is (x,1, y, y)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.py", line 206, in call
result = self.forward(*input, **kwargs)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/conv.py", line 237, in forward
self.padding, self.dilation, self.groups)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/functional.py", line 41, in conv2d
return f(input, weight, bias)
RuntimeError: CUDNN_STATUS_BAD_PARAM

how to modify the data generation script?

Whether it is encountered during the training process, the brightness of the output image is inconsistent with the input image.

Hi,

I just want to know the differences between checkpoint and pretrained model in pytorch. Does pytorch have checkpoint files like tensorflow? I just know the .pth file created by function of torch.save.( I am a new user in pytorch, please don't mind if my qusetion is easy...)

Thank you very much!

It seems that you do not transpose the data after reading from h5 file. The data in MATLAB is organized in (H, W, C, N). However, after reading it from h5 file in Python, it is organized in a reverse setting: (N, C, W, H). And PyTorch needs (N, C, H, W). So is it necessary to transpose the data? Thanks!

As I write, when we use very deep network, the output often be very big, that make loss nan,
why your network won't be nan?

Thanks.

Hi,

I just want to know how to get the best model from the checkpoint models. Because I have trained a series of checkpoints and use the last one checkpoint models (model_epoch_60.pth) to evaluate in the Set5 dataset. However, I can not achieve the performance as shown in this code.

Thank you very much!