Research map / Detection and segmentation / Open-set and few-shot detection
Towards More Practical Group Activity Detection: A New Benchmark and Model
Dongkeun Kim, Youngkil Song, Minsu Cho, Suha Kwak
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 group consistency loss (InfoNCE variant on actor embeddings) (ablation) | Grouping Transformer GAD model · Cafe split by view | Group mAP | Improved |
| keep grouping-attention layer in Grouping Transformer (ablation) | Grouping Transformer · Cafe split by view | Group mAP | Improved |
| keep self-attention and cross-attention layers (ablation) | Grouping Transformer · Cafe split by view | Group mAP | Improved |
| apply distance mask with threshold 0.2 on actor self-attention (ablation) | Grouping Transformer · Cafe split by view | Group mAP | Improved |
| use too small distance mask threshold 0.1 (ablation) | Grouping Transformer · Cafe split by view | Group mAP | Hurt |
| increase number of group tokens from 12 to 16 (ablation) | Grouping Transformer · Cafe split by view | Group mAP@1.0: 18.84 -> 15.03 | Hurt |
| match actors to groups in embedding space instead of 2D point matching (HGC) | Transformer GAD · Cafe split by view | Group mAP@1.0: 6.55 -> 18.84 | Improved |
| replace off-the-shelf clustering with learnable group tokens | GAD pipeline · Cafe inference time | inference time (s): 0.32 -> 0.10 | Improved |
| vary number of clusters for clustering-based methods (ablation) | ARG baseline · Cafe split by view | Group mAP@1.0: 11.03 (4) -> 1.27 (6) | Mixed |
| use detection-based tracklets instead of ground-truth tracklets | Grouping Transformer · Cafe split by view | Group mAP@1.0: 18.84 -> 14.36 | Hurt |
Closest papers by meaning
- Context-Interactive Reasoning for Group Activity Detection (ECCV 2026 · similarity 0.88)
- GroupFormer: Group Activity Recognition With Clustered Spatial-Temporal Transformer (ICCV 2021 · similarity 0.87)
- Joint Learning of Social Groups, Individuals Action and Sub-group Activities in Videos (ECCV 2020 · similarity 0.87)
- HiGCIN: Hierarchical Graph-Based Cross Inference Network for Group Activity Recognition (TPAMI 2020 · similarity 0.86)
- Learning Group Activity Features Through Person Attribute Prediction (CVPR 2024 · similarity 0.85)
- Bi-Causal: Group Activity Recognition via Bidirectional Causality (CVPR 2024 · similarity 0.84)
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.
