CUDA out of memory about srgan-pytorch

[Lornatang]

There may be multiple tasks running on the machine at the same time, or the machine may not have enough memory

[Lornatang]

What is the size of your image? Observe the fluctuation of the CUDA memory during the operation of the program.

[leminhhuy132]

When i calculate the psnr and update them i get the above error and my cuda memory as shown below but if i calculate the psnr but don't update them no more error.

[Lornatang]

The PSNR model is small and fast, it shouldn't take up so much memory, can you tell me the memory usage with and without PSNR?

[leminhhuy132]

When i comment psnr in train function, cuda memory as shown below

[Lornatang]

Does this issue occur during training or during testing?

[Lornatang]

train() or validate()？

[leminhhuy132]

only in train()

[Lornatang]

CUDA version?
CUDNN version?
PyTorch & Torchvision version?

[Lornatang]

In the current code, the psnr calculation will not be called, but will be called in validate(), check the difference between some native code and the latest code on Git?

[leminhhuy132]

CUDA 11.1
CUDNN 8.0.5
Pytorch 1.11.0+cu113
Torchvision 0.12.0+cu113
I run this in gg colab

[leminhhuy132]

yes, but i don't know why when i call train() it gives error

[Lornatang]

give me your config.py file. I check it

[leminhhuy132]

I edit your code in this repo, please help me check it
[email protected]:leminhhuy132/SRGAN_DACN.git

[Lornatang]

I edit your code in this repo, please help me check it [email protected]:leminhhuy132/SRGAN_DACN.git

in config.py 46 line batch_size 256? lt is so big?

[leminhhuy132]

i tried batch_size 16, it still got error