Jun-Yan Zhu
Jun-Yan Zhu
Could you share more details without us? You still have the issue even after you set ``-display_id 0``? Could you share the screenshot of your debugging output?
You can try fewshot vid2vid. If you have a smaller dataset, you need to reduce the network's capacity as well. It's hard to make any above methods work out of...
StyleGAN was not included in the paper. Training instruction will be added later.
$\hat z^s is the features encoded from the corresponding output image G(x), NOT a random image y.
@taesungp maybe we could fix it.
Please use a more recent torchvision.
Could you share with us the training and test command lines? Did you use the same flags (e.g., `-preprocess`)?
The model might overfit the training set. To prevent overfitting, you can either use a larger dataset or apply more aggressive augmentation (see the option `--preprocess` for more details.)
People find that normalizing data into zero-mean data helps model training. See this [post] for more [details](https://developers.google.com/machine-learning/data-prep/transform/normalization). Also, our generator's final layer has a TanH layer, which will only produce...
If you set `--input_nc 1 --output_nc 1`, it will require a single-channel image as input. For 1D data, you probably need to implement your own data loader, generator, and discriminator.