tonylins
commented
3 years ago Hi, we first jointly train the generator and discriminator (conventional GAN training), and then train the encoder for the fixed generator. Co-optimizing D, G, and E may bring some benefits, but it is beyond the scope of our paper.
Best, Ji
您好，感谢分享。 有一些不太理解的地方，希望能解答。
encoder, generator, discriminator的训练流程是怎样的？ 我猜测是 先discriminator, generator训练完成后， 使用generator来训练encoder。这种流程，encoder是不会影响generator。 那么是否可以三个模型一起训练。互相影响，达到最优。