search

una-dinosauria / human-motion-prediction

Simple baselines and RNNs for predicting human motion in tensorflow. Presented at CVPR 17.
MIT License
406 stars 141 forks source link

question about the srnn_loss #53

Open sukun1045 opened 5 years ago

sukun1045 commented 5 years ago

In the translate.py, when you test with srnn seeds, it looks like the srnn_loss you print out is only for the last action instead of an average loss of all actions.

una-dinosauria commented 5 years ago

Hi,

Sorry this is pretty hard for me to follow. Could you point to me to specific lines in the code, and what you think they should be instead? Better yet, you could open a pull request with the fix and then we can discuss it.

Cheers,

sukun1045 commented 5 years ago

Sure, in translate.py line 193, you have the srnn_loss for one action. However, after the for loop of all actions, in line 472, you print out the srnn_loss, which is only the srnn_loss for the last action. If i am correct, you should sum up srnn_losses for all actions and print out the average.

una-dinosauria commented 5 years ago

I think I see it. It works fine when there is only one action, but not when we train on all actions. Would you be up for submitting a PR?