action-recognition-pytorch icon indicating copy to clipboard operation
action-recognition-pytorch copied to clipboard

Published 20 hours ago • verified publisher icon IBM

Incorrect implementation of TSN

Open JiaMingLin opened this issue 2 years ago • 0 comments

Hi, thanks for this fantastic work. But I notice there is only forward implementation of TSN in this project, backward prop is missing. According to original paper, eq(5) in [L. Wang, TPAMI, 2019] , the gradient should be divided by "K", the number of segments.

[L. Wang, TPAMI, 2019] L. Wang, et al. "Temporal Segment Networks for Action Recognition in Videos", TPAMI VOL. 41, NO. 11 2019.

JiaMingLin avatar Apr 25 '22 03:04 JiaMingLin