Open yiskw713 opened 4 years ago
Joe Yue-Hei Ng et al.
WACV 2018
arXiv
Optical Flowの生成と行動認識を同時に行うモデルである ActionFlowNetを提案. 上図と下図の二つのモデルを提案しているが,上図の方が性能が良い.
精度的にはI3Dや,ImageNet-pretrained ResNetの方が高い.
一方でCAMを比較すると,提案手法の方がより動きの特徴に着目していることがわかる.
INFO
author
Joe Yue-Hei Ng et al.
affiliation
conference or year
WACV 2018
link
arXiv
概要
Optical Flowの生成と行動認識を同時に行うモデルである ActionFlowNetを提案. 上図と下図の二つのモデルを提案しているが,上図の方が性能が良い.
提案手法
検証
精度的にはI3Dや,ImageNet-pretrained ResNetの方が高い.![Screen Shot 2020-05-25 at 12 29 47](https://user-images.githubusercontent.com/38214459/82775825-d8988500-9e83-11ea-866a-509c7ca07d82.png)
一方でCAMを比較すると,提案手法の方がより動きの特徴に着目していることがわかる.![Screen Shot 2020-05-25 at 12 29 15](https://user-images.githubusercontent.com/38214459/82775827-da624880-9e83-11ea-93b0-613c7831a7c0.png)
date