Increased time for each epoch about shift-net_pytorch HOT 2 OPEN

Hi, Chaton,
we should have a version which just change one elements and not all of them.
I am not quite sure what this means.
B.T.W, I am back to work, will try to implement random_multi_shift this weak.

from shift-net_pytorch.

Hello @Zhaoyi-Yan ,

I timed each functions in the train.py because the training seemed to increase a lot. It appeared the function displaying the results on the web-page was increasing linearly with the number of images to display. It should just modify the last elements, making it o(1). Anyway, I removed those functions from the main loop. Also, I changed a bit the guidance loss implementation, making it more smooth to play with. I tried to increase the penalty, with a factor of 100. It seemed to give better results.
I am in vacation right now, but I am still running experiments.

Please, send me a notification when random_multi_shift is implemented, so I can have a look at it.

Best,
T.C

from shift-net_pytorch.