Jun-Yan Zhu
Jun-Yan Zhu
Unfortunately, we cannot find this model anymore. Maybe you can find a similar model in this [repo](https://github.com/AAnoosheh/ToDayGAN).
It depends on your application (whether you want to learn something simple or not, or whether your images are complex or not). You may want to use cropping during preprocessing...
You can crop the images during the training loop. We have implemented several options in `--preprocess'. See this [line](https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/14422fb8486a4a2bd991082c1cda50c3a41a755e/options/base_options.py#L49) for more details.
Here is a simpler implementation (in terms of the number of files): https://github.com/yunjey/mnist-svhn-transfer
I am not sure if we have this function `data_dependent_initialize` in our repo. Did you post it in the wrong repo?
Could you share more details?
Not sure. "Worse" means that your results on test data are worse than results on training data or validation data? One suggestion is to use the same augmentation/preprocessing steps across...
One sanity check is that you manually copy and paste the images from the test data to the validation data and see if there is a difference.
We don't have a Gradio/HF GUI as the model can be used in many different applications, each of which requires a separate UI. People have built GUI for sketch2photo applications,...
it might be related to data loading. If you use our datasets, have you downloaded the data? If you use your own datasets, you may want to check whether the...