search

alibaba-mmai-research / TAdaConv

[ICLR 2022] TAda! Temporally-Adaptive Convolutions for Video Understanding. This codebase provides solutions for video classification, video representation learning and temporal detection.
https://tadaconv-iclr2022.github.io
Apache License 2.0
225 stars 30 forks source link

How to use R(2+1)D with TAda? #9

Closed sja9527 closed 2 years ago

sja9527 commented 2 years ago

Can you show us the detail of using R(2+1)D with TAdaConv?

huang-ziyuan commented 2 years ago

Hi, we just updated the implementation for TAdaConv3d.

For using R(2+1)D with TAdaConv,

  1. follow the guideline in README.md to replace the 2d conv with TAdaConv2d
  2. similarly, replace the original 1d conv with TAdaConv3d (with kernel set to [3,1,1]).

For the implementation of the R(2+1)D in our paper, please refer to r2plus1d_branch.py.