search

dyelax / Adversarial_Video_Generation

A TensorFlow Implementation of "Deep Multi-Scale Video Prediction Beyond Mean Square Error" by Mathieu, Couprie & LeCun.
MIT License
736 stars 184 forks source link

multi-scale architecture is to learn difference #11

Open zachluo opened 7 years ago

zachluo commented 7 years ago

hi,dyelax I found that in the original paper, each scale network is to learn Yk - uk(Yk-1). But in your implementation, it is to learn Yk. Did I misunderstand ?

dyelax commented 7 years ago

I'm not sure I understand your question. Can you point to where in the paper and where in my code you are referring to?

zachluo commented 7 years ago

@dyelax hi, I mean in the paper, there is an equation (2) , which means Generator Network at size k is to learn the difference and it should add the previous size k/2 output. In your code, it should add some codes like this before g_model.py:271 : if last_gen_frames is not None: preds += last_gen_frames return preds, scale_gts Am i right ?