storymy / take-off-eyeglasses Goto Github PK

Thanks for your excellent work. I have a little question. Whether the real data CelebA (with eyeglasses) is used only for domain adaptation to uniform real domain and synthetic domain? Or it is used in training, for example, mixup real image and synthetic image for training, or fine tune with real image, I have not find this question in the paper, so sorry to bother you.

This work is interesting and well-done! The inference code is clear, but I have trouble reproducing the traning phase. I wanna ask several questions.

1. Could you release the training code?
2. I follow all the settings avaliable in the paper to reproduce, but the inference results are excursive. Could you provide more clear settings in the implementation process.
3. Could you provide the evaluation code for fair comparison.

Really thanks for your prompt reply. Thanks again!

line54：transforms.Resize(args.img_size)

This line of code will scaling the picture to 256 * 256.However, if I input a 296 * 294 picture, the following error will appear
"RuntimeError: Sizes of tensors must match except in dimension 2. Got 296 and 294 (The offending index is 0)"

after I manually scaled the picture to 256 * 256, the program can run correctly and output the results.

我正在尝试复现您的论文并且尝试进行一些可能的改进，如果可以，能提供一下训练过程的代码吗

你好，我想问一下，为什么会出现这个问题呀?
Hello, I'd like to ask, why is this problem?

Traceback (most recent call last):
File "easy_use.py", line 97, in
generate_results(args, device)
File "easy_use.py", line 66, in generate_results
img = transform(img)
File "/disk/sdb/public_user/anaconda3/envs/wxy_env/lib/python3.8/site-packages/torchvision/transforms/transforms.py", line 67, in call
img = t(img)
File "/disk/sdb/public_user/anaconda3/envs/wxy_env/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in call_impl
result = self.forward(*input, **kwargs)
File "/disk/sdb/public_user/anaconda3/envs/wxy_env/lib/python3.8/site-packages/torchvision/transforms/transforms.py", line 226, in forward
return F.normalize(tensor, self.mean, self.std, self.inplace)
File "/disk/sdb/public_user/anaconda3/envs/wxy_env/lib/python3.8/site-packages/torchvision/transforms/functional.py", line 284, in normalize
tensor.sub(mean).div_(std)
RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 0

Lol. Handing this L to myself. The replicate.com demo just seems to change the input size from 256 to 512 which drastically improves the quality. Thanks for the AWESOME research dude!