Zhaoyi-Yan

Change here https://github.com/Zhaoyi-Yan/Shift-Net_pytorch/blob/master/models/modules/shift_unet.py#L41. Relapce some of them with "residual blocks", it should be weight-lighter than unet-skip blocks, as unet-skip block is feature-concatenating while residual-block is feature-adding.

Hi, Chaton, `we should have a version which just change one elements and not all of them.` I am not quite sure what this means. B.T.W, I am back to...

Hi, you should have known the answer. If not, I will explain.

Hmm, I think several problems may exist: 1. Your dataset is too small. You can use 'RandomResizeCrop', 'flipping', etc as your data augmentation methods. 2. You should adjust https://github.com/Zhaoyi-Yan/Shift-Net/blob/master/train.lua#L102 `local...

You can post the error information

Put all your training images to the custom path, then change `dataroot` to your custom path in `https://github.com/Zhaoyi-Yan/Shift-Net_pytorch/blob/master/options/base_options.py`. I have updated `Readme` here: https://github.com/Zhaoyi-Yan/Shift-Net_pytorch#train-models

Setting `--gpu_ids='-1'` will work.

Hi, it is pytorch, not Torch...

`python train.py --which_model_netG='unet_shift_triple' --model='shiftnet' --shift_sz=1 --mask_thred=1 --gpu_ids='-1' `works fine for me.

Change some values here : https://github.com/Zhaoyi-Yan/Shift-Net_pytorch/blob/master/util/util.py#L121 @youarenotaloneor