about the pretrained model

[yuffon]

The flow based model is very very interesting. But it is too GPU hungry. I see that some pretrained models are uploaded. Now I have a question about the pretrained model on CelebA. For unoptimized model, the input for encoding includes three tensors: 'enc_x': 'input/image', 'enc_x_d': 'input/downsampled_image', 'enc_y': 'input/label' I want to know the relationship between enc_x and enc_x_d. If I want to test the model on 32x32 CelebA, how should I use this pretrained model?

[isharifi]

That's also my question.