Research map / Video / Action recognition
Actor-Transformers for Group Activity Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 positional encoding of actor locations (ablation) | Pose Actor-Transformer · Volleyball | group activity accuracy: 91.0 -> 92.3 | Helped |
| increase attention heads from 1 to 2 (ablation) | Pose Actor-Transformer · Volleyball | group activity accuracy: 92.3 -> 91.4 | Hurt |
| increase encoder layers from 1 to 2 (ablation) | Pose Actor-Transformer · Volleyball | group activity accuracy: 92.3 -> 92.1 | No clear effect |
| replace graph aggregation with actor-transformer aggregation (ablation) | actor feature aggregation · Volleyball, Pose | group activity accuracy: 92.0 -> 92.3 | Helped |
| replace graph aggregation with actor-transformer aggregation (ablation) | actor feature aggregation · Volleyball, RGB | group activity accuracy: 91.1 -> 91.4 | Helped |
| replace graph aggregation with actor-transformer aggregation (ablation) | actor feature aggregation · Volleyball, Flow | group activity accuracy: 89.5 -> 91.5 | Helped |
| use late rather than early summation fusion of Pose and RGB (ablation) | Actor-Transformer · Volleyball, Pose + RGB | group activity accuracy: 91.2 -> 93.5 | Helped |
| use late rather than early concatenation fusion of Pose and RGB (ablation) | Actor-Transformer · Volleyball, Pose + RGB | group activity accuracy: 91.8 -> 93.5 | Helped |
| use late rather than early summation fusion of Pose and Flow (ablation) | Actor-Transformer · Volleyball, Pose + Flow | group activity accuracy: 88.5 -> 94.4 | Helped |
| use late rather than early concatenation fusion of Pose and Flow (ablation) | Actor-Transformer · Volleyball, Pose + Flow | group activity accuracy: 89.7 -> 94.4 | Helped |
| combine static pose features with a dynamic representation | Actor-Transformer · Volleyball | group activity accuracy: 91.4 -> 94.4 | Helped |
| combine RGB and Flow representations | Actor-Transformer · Collective | group activity accuracy: 90.8 -> 92.8 | Helped |
Closest papers by meaning
- Learning Actor Relation Graphs for Group Activity Recognition (CVPR 2019 · similarity 0.92)
- Dual-AI: Dual-Path Actor Interaction Learning for Group Activity Recognition (CVPR 2022 · similarity 0.90)
- Detector-Free Weakly Supervised Group Activity Recognition (CVPR 2022 · similarity 0.90)
- Learning Visual Context for Group Activity Recognition (AAAI 2021 · similarity 0.88)
- Learning from Synthetic Human Group Activities (CVPR 2024 · similarity 0.85)
- Learning Group Actions on Latent Representations (NeurIPS 2024 · similarity 0.85)
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.