AI Paper Atlas

Research map / Video / Action recognition

Multi-Modal Multi-Action Video Recognition

Zhensheng Shi, Ju Liang, Qianqian Li, Haiyong Zheng, Zhaorui Gu, Junyu Dong, Bing Zheng

ICCV 2021 · Video › Action recognition · approach: Disentangled action recognition · 14 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace fully-connected classifier with visual GCN (ablation)R(2+1)D-34 · Mini M-MiTmAP: 54.8 -> 55.0Improved
replace fully-connected classifier with visual GCN (ablation)R(2+1)D-34 · Mini M-MiTtop-1 accuracy: 52.1 -> 53.3Improved
add audio GCN to visual model (ablation)J(H, Gv) · Mini M-MiTmAP: 55.0 -> 58.0Improved
add textual GCN to visual model (ablation)J(H, Gv) · Mini M-MiTmAP: 55.0 -> 58.2Improved
add visual GCN to audio GCN joint model (ablation)J(H, Ga) · Mini M-MiTmAP: 58.0 -> 58.2Improved
add visual GCN to textual GCN joint model (ablation)J(H, Gt) · Mini M-MiTmAP: 58.2 -> 58.5Improved
add textual GCN to visual plus audio GCN (all three modalities) (ablation)J(H, Gv, Ga) · Mini M-MiTmAP: 58.2 -> 58.7Improved
add third modality over two modalities (ablation)J(H, Gv, Ga, Gt) · Mini M-MiTtop-1 accuracy: 55.1 -> 55.0No clear effect
use 2 synonymous audio features per action instead of 1 (ablation)J(H, Ga) · Mini M-MiTmAP: 58.0 -> 57.8Hurt
use 3 synonymous audio features per action instead of 1 (ablation)J(H, Ga) · Mini M-MiTmAP: 58.0 -> 57.4Hurt
BERT 768D instead of GloVe 300D text lexicon (ablation)J(H, Gt) · Mini M-MiTmAP: 58.2 -> 58.2No clear effect
merge audio and textual lexicons into one audio-textual GCN (ablation)J(H, Ga,t) · Mini M-MiTmAP: 58.0 -> 58.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy