Research map / Video / Action recognition
Context-Interactive Reasoning for Group Activity Detection
Xi Ai, Weihong Ren, Shuhuan Han, Haoran Xu, Qian Dong, Pengyang Su, Zijian Wang, Shengchun Lin, Zhiyong Wang, Honghai Liu
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 STIR and LRCC to Practical GAD baseline (ablation) | Practical GAD + DINOv2 · Cafe split-by-place | Group mAP1.0: 10.45 -> 23.82 | Improved |
| add STIR (soft topology attention) (ablation) | baseline DINOv2 · Cafe split-by-place | Group mAP1.0: 10.45 -> 15.69 | Improved |
| add LRCC (gated scene context) (ablation) | baseline DINOv2 · Cafe split-by-place | Group mAP1.0: 10.45 -> 18.98 | Improved |
| add STIR to LRCC model (ablation) | LRCC-only DINOv2 · Cafe split-by-place | Group mAP0.5: 35.58 -> 47.97 | Improved |
| add LRCC to STIR model (ablation) | STIR-only DINOv2 · Cafe split-by-place | Group mAP0.5: 35.16 -> 47.97 | Improved |
| soft learnable distance penalty plus bias instead of hard distance mask (ablation) | STIR · Cafe split-by-place | Group mAP1.0: 20.70 -> 23.82 | Improved |
| full STIR instead of spatial bias only (ablation) | STIR · Cafe split-by-place | Group mAP0.5: 45.10 -> 47.97 | Improved |
| token-type-specific two-branch gates instead of shared gate (ablation) | LRCC · Cafe split-by-place | Group mAP1.0: 16.45 -> 23.82 | Improved |
| two-branch gating instead of static global pooling (ablation) | LRCC · Cafe split-by-place | Group mAP0.5: 43.28 -> 47.97 | Improved |
| replace ResNet-18 with DINOv2 backbone in baseline (ablation) | Practical GAD baseline · Cafe split-by-place | Group mAP1.0: 8.75 -> 10.45 | Mixed |
| add concat or add context fusion instead of gating (ablation) | LRCC · Cafe split-by-place | Outlier mIoU: 69.71 -> 70.36 (concat), 68.21 (add) | Mixed |
| use proposed method instead of ProGraD (full FT) | ProGraD · Cafe split-by-place | Group mAP1.0: 20.42 -> 23.82 | Improved |
Closest papers by meaning
- Learning Visual Context for Group Activity Recognition (AAAI 2021 · similarity 0.90)
- Detector-Free Weakly Supervised Group Activity Recognition (CVPR 2022 · similarity 0.90)
- Joint Learning of Social Groups, Individuals Action and Sub-group Activities in Videos (ECCV 2020 · similarity 0.89)
- HiGCIN: Hierarchical Graph-Based Cross Inference Network for Group Activity Recognition (TPAMI 2020 · similarity 0.88)
- GroupFormer: Group Activity Recognition With Clustered Spatial-Temporal Transformer (ICCV 2021 · similarity 0.88)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · 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.
