search

majedelhelou / SFM

(ECCV 2020) Stochastic Frequency Masking to Improve Super-Resolution and Denoising Networks
153 stars 22 forks source link

update link of pretrained model #1

Open danielkaifeng opened 4 years ago

danielkaifeng commented 4 years ago

hi, very nice job on SR in frequency domain. Could you update the pretrained model in gdrive or baidu netdisk? Thanks!

sfm-sr-denoising commented 4 years ago

Thank you for your positive comment! All our pre-trained models for denoising are uploaded on this repository, but for SR the models are larger and could not be uploaded here. We still cannot upload them to GDrive or Baidu NetDisk yet, because the paper is under submission and this repository has to remain anonymous. We will upload these models and update the link as soon as possible. In the meantime, all our networks can be obtained again by following the re-training code and guidelines. Thank you for your patience.

danielkaifeng commented 4 years ago

No pro, thanks for your work!

buildist commented 4 years ago

Hi, I was interested in comparing your method to ESRGAN-FS but I haven't been able to run the training code, it seems like loss and mseloss here are both undefined. Is something missing? https://github.com/sfm-sr-denoising/sfm/blob/master/SR/train.py#L110

sfm-sr-denoising commented 4 years ago

Hi, I was interested in comparing your method to ESRGAN-FS but I haven't been able to run the training code, it seems like loss and mseloss here are both undefined. Is something missing? https://github.com/sfm-sr-denoising/sfm/blob/master/SR/train.py#L110

Hi, two lines were accidentally deleted, adding these two lines should work:

loss = L1criterion(output, x_data)
mseloss = MSEcriterion(output, x_data)

The code is updated