Research map / Graph learning / Knowledge and temporal graphs
Learning Actor Relation Graphs 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 actor relation graph reasoning (ablation) | base model · Volleyball | group activity accuracy: 89.8% -> 91.3% | Helped |
| Use dot-product or embedded dot-product rather than relation network for appearance relation (ablation) | single-frame ARG · Volleyball | group activity accuracy: 90.7% -> 91.3% | Helped |
| Add position relation (ablation) | single-frame ARG with embedded dot-product · Volleyball | group activity accuracy: 91.3% -> 91.6% | Helped |
| Use distance mask rather than distance encoding (ablation) | ARG · Volleyball | group activity accuracy: 91.5% -> 91.6% | No clear effect |
| Increase relation graphs from 1 to multiple graphs (ablation) | ARG · Volleyball | group activity accuracy: 91.6% -> 92.1% | Helped |
| Fuse graph outputs after graph convolution by summation rather than fusing graphs before convolution (ablation) | 16-graph ARG · Volleyball | group activity accuracy: 90.8% -> 92.1% | Helped |
| Use late fusion by summation rather than concatenation (ablation) | 16-graph ARG · Volleyball | group activity accuracy: 91.9% -> 92.1% | No clear effect |
| Add 3-frame TSN aggregation (ablation) | ARG · Volleyball | group activity accuracy: 92.1% -> 92.3% | Helped |
| Build temporal graphs over 3 frames rather than aggregate frame predictions with TSN (ablation) | ARG · Volleyball | group activity accuracy: 92.3% -> 92.5% | Helped |
| Use ARG with Inception-v3 and ground-truth boxes rather than SSU with the same backbone and box type | Volleyball · Volleyball | group activity accuracy: 90.6% -> 92.5% | Helped |
| Use ARG with Inception-v3 and ground-truth boxes rather than SSU with the same backbone and box type | Volleyball · Volleyball | individual action accuracy: 81.8% -> 83.0% | Helped |
| Use ARG with Inception-v3 and ground-truth boxes rather than stagNet with ground-truth boxes | group activity recognition · Collective Activity | group activity accuracy: 89.1% -> 91.0% | Helped |
Closest papers by meaning
- Actor-Transformers for Group Activity Recognition (CVPR 2020 · similarity 0.92)
- Progressive Relation Learning for Group Activity Recognition (CVPR 2020 · similarity 0.91)
- Convolutional Relational Machine for Group Activity Recognition (CVPR 2019 · similarity 0.91)
- Dual-AI: Dual-Path Actor Interaction Learning for Group Activity Recognition (CVPR 2022 · similarity 0.90)
- Structure Inference Machines: Recurrent Neural Networks for Analyzing Relations in Group Activity Recognition (CVPR 2016 · similarity 0.88)
- 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.