search

eriklindernoren / PyTorch-GAN

PyTorch implementations of Generative Adversarial Networks.
MIT License
16.22k stars 4.05k forks source link

Info GAN training mechanism #131

Open zwvews opened 3 years ago

zwvews commented 3 years ago

In this implementation, the information loss is trained to simultaneously optimize G and D which is not True. Based on my understanding and the OpenAI implementation https://github.com/openai/InfoGAN, the information loss should be trained similarly as the GAN loss.

ariel415el commented 3 years ago

Also, they are using the GT labels of MNIST which up to my knowledge not what they are supposed to do