cryu854

Upgrade the version of qpth might be work. my packages: Python3.7.6 + Pytorch 1.2.0 + qpth 0.015

It is difficult to get optimizer compatible with tf 1.x and tf 2.x at the same time, since `tf.train.Optimizer` is deprecated in tf 2.x, but I will try to implement...

Sorry for the late reply. Please use tensorflow 2 or use the cpu training for now, I will start investigating this issue soon and note you once I have any...

Hi @hjzzju I didn't calculate the FID score for CUT unfortunately, the reason is that when calculating FID, we need more than 2048 samples(because inception_v3's output has 2048 dims) to...

Only at training phase needs to specify the image size because it uses a VGG network to extract features. After training is complete, you can transfer images of arbitrary size....

Hi @moldach , you can use the following command to transfer a content image ``` python main.py evaluate --content ./path/to/content/image.jpg \ --weights ./path/to/weights \ --result ./path/to/save/results/image.jpg ```

Once you have trained the model based on your style image, you will get a new model named `imagenet-vgg-verydeep-19.mat` for your case. If you want to stylize the content image,...

@Donglin-Wang2 Since the trick was not mentioned in their paper, you can refer to their official configuration [tamarott/SinGAN/config.py](https://github.com/tamarott/SinGAN/blob/master/config.py). But it seems to me that they just some hyperparameters to play...