Research map / Video / Action recognition
Social Adaptive Module for Weakly-supervised Group Activity Recognition
Rui Yan, Lingxi Xie, Jinhui Tang, Xiangbo Shu, Qi Tian
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Spatial-SAM selecting 8 of 14 proposals (pruning) (ablation) | Ours baseline · NBA | accuracy (%): 44.6 -> 50.3 | Improved |
| add Spatial-SAM relational embedding without pruning (14 proposals) (ablation) | Ours baseline · NBA | accuracy (%): 44.6 -> 46.8 | Improved |
| add pruning of proposals (Kp=8 from Np=14) on top of unpruned SAM (ablation) | Spatial-SAM · NBA | accuracy (%): 46.8 -> 50.3 | Improved |
| add Temporal-SAM to Spatial-SAM (ablation) | Spatial-SAM (B3) · NBA | mean accuracy (%): 43.6 -> 47.5 | Improved |
| add Temporal-SAM to Spatial-SAM (ablation) | Spatial-SAM (B3) · NBA | accuracy (%): 50.3 -> 49.1 | Hurt |
| use probability-aware instead of quantity-aware proposal count (ablation) | Spatial-SAM (Kp=8) · NBA | mean accuracy (%): 43.6 -> 42.6 | Hurt |
| add Spatial-SAM with pruning to the base model (ablation) | Ours w/o SAM · NBA | mean accuracy (%): 39.5 -> 43.6 | Improved |
| add Spatial-SAM with pruning (Np=16, Kp=12) (ablation) | Ours baseline · Volleyball Dataset (weak supervision) | accuracy (%): 91.5 -> 93.1 | Improved |
| add SAM relational embedding without pruning (Np=16, Kp=16) (ablation) | Ours baseline · Volleyball Dataset (weak supervision) | accuracy (%): 91.5 -> 92.4 | Improved |
| use fewer input proposals without pruning (Np=12, Kp=12) (ablation) | Spatial-SAM · Volleyball Dataset (weak supervision) | accuracy (%): 91.5 -> 91.9 | No clear effect |
| train ARG with only video-level labels instead of full supervision | ARG · Volleyball Dataset | accuracy (%): 94.0 -> 90.7 | Hurt |
| switch SAM backbone from ResNet-18 to Inception-v3 | Ours (weakly supervised) · Volleyball Dataset | accuracy (%): 93.1 -> 94.0 | Improved |
Closest papers by meaning
- Detector-Free Weakly Supervised Group Activity Recognition (CVPR 2022 · similarity 0.91)
- GroupFormer: Group Activity Recognition With Clustered Spatial-Temporal Transformer (ICCV 2021 · similarity 0.89)
- Joint Learning of Social Groups, Individuals Action and Sub-group Activities in Videos (ECCV 2020 · similarity 0.88)
- Hunting Group Clues with Transformers for Social Group Activity Recognition (ECCV 2022 · similarity 0.88)
- A Hierarchical Deep Temporal Model for Group Activity Recognition (CVPR 2016 · similarity 0.87)
- Learning Group Activity Features Through Person Attribute Prediction (CVPR 2024 · similarity 0.86)
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.
