search

saikatdutta / Stacked_DMSHN_bokeh

Official Repository for our CVPRW (MAI'21) paper.
MIT License
23 stars 9 forks source link

train #7

Closed 13sunmin closed 2 years ago

13sunmin commented 2 years ago

Hello Sir,

While trying to replicate your work, I'm having some difficulties in training the model. The Adam optimizer is returning nan, and am not sure if its due to the wrong implementation of the loss function (tried implementing after reading your paper). It would be greatly helpful if you could share your training script. You could send it to my email id : 849972256@qq.com

Thank you.

saikatdutta commented 2 years ago

Hi,

Thanks for showing interest in our work. We've lost the training script, so I won't be able to share the same. However, I can advise you the following:

  1. Make sure you are using the correct pytorch version.
  2. We have linked the loss function repositories in "Useful Repositories" (check Readme).
  3. We used an older version of MS-SSIM. Please undo this change: https://github.com/jorge-pessoa/pytorch-msssim/commit/ff882263779e94c29ec4e76a432d2ec80df8796b#diff-b583e4c6807ec522fa88e389e2245135a1686c73bbcdfa8baa8a7068288cf49f
  4. You can start by training with a simple loss function (L1) to check if there is some other problem.

Let me know if these fixes work.

saikatdutta commented 2 years ago

Closing. Feel free to open if needed.