search

kamenbliznashki / pixel_models

Pytorch implementations of autoregressive pixel models - PixelCNN, PixelCNN++, PixelSNAIL
35 stars 8 forks source link

conditional sampling #2

Open Dashi-1997 opened 4 years ago

Dashi-1997 commented 4 years ago

Hello, Can your code implement conditional sampling?

kamenbliznashki commented 4 years ago

Yes, all the models can be conditional. I used conditional PixelCNN and PixelSNAIL in the VQVAE2 implementation in my generative models repo here https://github.com/kamenbliznashki/generative_models, where the pixel models are used as priors and sampled from conditionally. Take a look at vqvae_prior.py

On Thu, Jul 2, 2020 at 7:44 AM Dashi-1997 notifications@github.com wrote:

Hello, Can your code implement conditional sampling?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kamenbliznashki/pixel_models/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG3JPP7JOSXEW5TKF5UNLWTRZRXLFANCNFSM4OOZXMIQ .

Dashi-1997 commented 4 years ago

Thanks for your reply! I am wondering about how to use conditions in VQ-VAE and I will read your code carefully.

liyikang0613 commented 3 years ago

Hi, Can you please tell me how to train the Pixel CNN++ on other datasets besides the three datasets you `provided? This is my set data function but I have no idea how to add it to the model. Could you please help me with it? Thank you. setdata.zip

kamenbliznashki commented 3 years ago

Hi - the best way to include new datasets is to update fetch_dataloaders in main.py for the dataset and dataloaders you'd like to train on.