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

Loss weighting #32

Open mrgloom opened 5 years ago

mrgloom commented 5 years ago

As I can see here: https://github.com/dyelax/Adversarial_Video_Generation/blob/9b0d10f357871231bbec06c610363588812216e1/Code/g_model.py#L189 https://github.com/dyelax/Adversarial_Video_Generation/blob/9b0d10f357871231bbec06c610363588812216e1/Code/loss_functions.py#L7 there is no weighting on GDL loss, but as I understand absolute values of L2 loss and GDL loss will have large difference. Is it normal?

@dyelax

mrgloom commented 5 years ago

Here I also see weights for losses, but looks like they are not used: https://github.com/dyelax/Adversarial_Video_Generation/blob/9b0d10f357871231bbec06c610363588812216e1/Code/constants.py#L156