TengdaHan / DPC

Video Representation Learning by Dense Predictive Coding. Tengda Han, Weidi Xie, Andrew Zisserman.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Reproducing the paper's results on HMDB51/UCF101

mengyuest opened this issue · comments

Hi, I wonder which model (and setting) can we use to reproduce the result for HMDB51/UCF101? I used model 3D-ResNet18-Kinetics400-128x128 but only got 1% top-1 accuracy on HMDB51. Thanks for your help.

Hi, the weights provided now is the self-supervised weights, i.e. the weights haven't seen any action labels.

To reproduce, you can follow the instruction here with the pretrained weights, and finetune the classifier on UCF101 or HMDB51.

Email me if you have more questions.

Issue has been solved