Research map / Video / Action recognition
GroupFormer: Group Activity Recognition With Clustered Spatial-Temporal Transformer
Shuaicheng Li, Qianggang Cao, Lingbo Liu, Kunlin Yang, Shinan Liu, Jun Hou, Shuai Yi
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 the full CSTT (cross-decoder spatial-temporal transformer) instead of an FC layer (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 91.0 -> 94.1 | Improved |
| add cross individual decoders to parallel encoders (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 92.2 -> 94.1 | Improved |
| model spatial and temporal context in parallel rather than spatial only (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 91.8 -> 92.2 | Improved |
| stack spatial and temporal encoders versus parallel summation (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 92.2 -> 92.6 | Improved |
| add clustered attention (4 clusters) to the spatial-temporal transformer (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 93.4 -> 94.1 | Improved |
| increase clusters from 4 to 6 (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 94.1 -> 93.4 | Hurt |
| add intra-group attention only (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 93.4 -> 93.8 | Improved |
| add inter-group attention only (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 93.4 -> 93.6 | Improved |
| combine intra- and inter-group attention (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 93.4 -> 94.1 | Improved |
| use one CSTT block instead of FC layer (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 91.0 -> 93.6 | Improved |
| stack 3 CSTT blocks instead of 1 (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 93.6 -> 94.1 | Improved |
| stack 4 CSTT blocks instead of 3 (ablation) | GroupFormer · Volleyball val (Inception-v3) | group activity accuracy: 94.1 -> 93.9 | Hurt |
Later papers that compared with this method
1 later accepted paper reports GroupFormer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- COMPOSER: Compositional Reasoning of Group Activity in Videos with Keypoint-Only Modality (ECCV 2022 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Hunting Group Clues with Transformers for Social Group Activity Recognition (ECCV 2022 · similarity 0.92)
- Detector-Free Weakly Supervised Group Activity Recognition (CVPR 2022 · similarity 0.92)
- Design and Analysis of Efficient Attention in Transformers for Social Group Activity Recognition (IJCV 2024 · similarity 0.91)
- Joint Learning of Social Groups, Individuals Action and Sub-group Activities in Videos (ECCV 2020 · similarity 0.91)
- Actor-Transformers for Group Activity Recognition (CVPR 2020 · similarity 0.90)
- Bi-Causal: Group Activity Recognition via Bidirectional Causality (CVPR 2024 · similarity 0.87)
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.
