Research map / Video / Action recognition
SBGAR: Semantics Based Group Activity Recognition
Xin Li, Mooi Choo Chuah
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 optical flow images to RGB frames (ablation) | SBGAR · Collective Activity Dataset | accuracy (%): 83.7 -> 86.1 | Improved |
| add optical flow images to RGB frames (ablation) | SBGAR · Volleyball Dataset | accuracy (%): 54.3 -> 66.9 | Improved |
| add RGB frames to optical flow images (ablation) | SBGAR · Collective Activity Dataset | accuracy (%): 70.1 -> 86.1 | Improved |
| add RGB frames to optical flow images (ablation) | SBGAR · Volleyball Dataset | accuracy (%): 38.7 -> 66.9 | Improved |
| use a temporal LSTM over multiple frames instead of single-frame classification | Inception-v3 image features · Collective Activity Dataset | accuracy (%): 67.2 -> 68.5 | Improved |
| use a temporal LSTM over multiple frames instead of single-frame classification | Inception-v3 image features · Volleyball Dataset | accuracy (%): 41.9 -> 44.3 | Improved |
| use caption-based semantic LSTM instead of direct image features LSTM | temporal model · Volleyball Dataset | accuracy (%): 44.3 -> 66.9 | Improved |
| use frames both before and after the current frame instead of before only (Before(10) to Before(10)After(10)) (ablation) | SBGAR · Collective Activity Dataset | accuracy (%): 85.7 -> 86.4 | Improved |
| use frames both before and after the current frame instead of before only (Before(10) to Before(10)After(10)) (ablation) | SBGAR · Volleyball Dataset | accuracy (%): 64.7 -> 67.7 | Improved |
| use only past frames Before(10) versus only future frames After(10) (ablation) | SBGAR · Collective Activity vs Volleyball | accuracy (%): 85.7 vs 84.7 on Collective; 64.7 vs 65.1 on Volleyball | Mixed |
| train LSTM1 for more epochs (stable beyond 500) (ablation) | SBGAR · Collective and Volleyball | accuracy | Improved |
| train LSTM2 for more epochs (stable after 200) (ablation) | SBGAR · Collective and Volleyball | accuracy | Improved |
Closest papers by meaning
- GroupFormer: Group Activity Recognition With Clustered Spatial-Temporal Transformer (ICCV 2021 · similarity 0.89)
- A Hierarchical Deep Temporal Model for Group Activity Recognition (CVPR 2016 · similarity 0.88)
- Detector-Free Weakly Supervised Group Activity Recognition (CVPR 2022 · similarity 0.88)
- stagNet: An Attentive Semantic RNN for Group Activity Recognition (ECCV 2018 · similarity 0.88)
- Hunting Group Clues with Transformers for Social Group Activity Recognition (ECCV 2022 · similarity 0.87)
- Social Adaptive Module for Weakly-supervised Group Activity Recognition (ECCV 2020 · 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.
