search

rubenvillegas / iclr2017mcnet

Tensorflow implementation of the ICLR 2017 paper: Decomposing Motion and Content for Natural Video Sequence Prediction
https://sites.google.com/a/umich.edu/rubenevillegas/iclr2017
MIT License
109 stars 34 forks source link

gdl loss #5

Closed kikyou123 closed 7 years ago

kikyou123 commented 7 years ago

what the UCF-101 dataset ？

rubenvillegas commented 7 years ago

Did you figure out the reason of the shapes in the first issue? It seems like it disappeared from the thread. I don't understand the second question.

kikyou123 commented 7 years ago

the first issure i have solved. the second issure i am asking why the train script don not have UCF-101 datasets?

rubenvillegas commented 7 years ago

Because the qualitative results shown in the paper are from training on Sports 1M and testing on UCF101. We found that numerically it gives very similar results, thus the generalizability of our method.

kikyou123 commented 7 years ago

thank you for your replay. i understand it