Research map / Video / Action recognition
Something-Else: Compositional Action Recognition With Spatial-Temporal Interaction Networks
Joanna Materzynska, Tete Xiao, Roei Herzig, Huijuan Xu, Xiaolong Wang, Trevor Darrell
Check 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 object identity embeddings to STIN (ablation) | STIN · Something-Else compositional split with ground-truth boxes | top-1 accuracy: 47.1 -> 51.3 | Improved |
| Add object identity embeddings to STIN (ablation) | STIN · Something-Else compositional split with detected boxes | top-1 accuracy: 34.1 -> 36.7 | Improved |
| Add non-local aggregation to STIN with OIE (ablation) | STIN + OIE · Something-Else compositional split with ground-truth boxes | top-1 accuracy: 51.3 -> 51.4 | No clear effect |
| Add non-local aggregation to STIN with OIE (ablation) | STIN + OIE · Something-Else compositional split with detected boxes | top-1 accuracy: 36.7 -> 37.2 | Improved |
| Add object identity embeddings to STIN (ablation) | STIN · Something-Else few-shot novel categories, ground-truth boxes, 5-shot | accuracy: 24.5 -> 25.8 | Improved |
| Add non-local aggregation to STIN with OIE (ablation) | STIN + OIE · Something-Else few-shot novel categories, ground-truth boxes, 5-shot | accuracy: 25.8 -> 27.7 | Improved |
| Add object identity embeddings to STIN (ablation) | STIN · Something-Else few-shot novel categories, detected boxes, 5-shot | accuracy: 14.2 -> 16.3 | Improved |
| Add non-local aggregation to STIN with OIE (ablation) | STIN + OIE · Something-Else few-shot novel categories, detected boxes, 5-shot | accuracy: 16.3 -> 17.7 | Improved |
| Train on only one object category, box (ablation) | STIN + OIE · Something-Else one-class compositional split with ground-truth boxes | top-1 accuracy: 14.7 -> 28.5 | Improved |
| Combine I3D appearance features with STIN interaction features | I3D · Something-Something V2 original split | top-1 accuracy: 55.5 -> 60.2 | Improved |
| Use an ensemble of I3D and STIN rather than I3D alone | I3D · Something-Else compositional split with ground-truth boxes | top-1 accuracy: 46.8 -> 58.1 | Improved |
| Use an ensemble of STRG and STIN | STRG · Something-Else compositional split with detected boxes | top-1 accuracy: 52.3 -> 56.2 | Improved |
Closest papers by meaning
- Compositional Learning for Human Object Interaction (ECCV 2018 · similarity 0.90)
- Progressive Instance-Aware Feature Learning for Compositional Action Recognition (TPAMI 2023 · similarity 0.89)
- Interaction Compass: Multi-Label Zero-Shot Learning of Human-Object Interactions via Spatial Relations (ICCV 2021 · similarity 0.89)
- C2C: Component-to-Composition Learning for Zero-Shot Compositional Action Recognition (ECCV 2024 · similarity 0.89)
- Object Level Visual Reasoning in Videos (ECCV 2018 · similarity 0.89)
- VELOCITI: Benchmarking Video-Language Compositional Reasoning with Strict Entailment (CVPR 2025 · 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.
