Research map / Video / Action recognition
Unified Graph Structured Models for Video Understanding
Anurag Arnab, Chen Sun, Cordelia Schmid
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 spatial message passing with GAT (ablation) | SlowFast R50 · Action Genome SGCls | Recall@20: 48.9 -> 51.1 | Improved |
| use GAT instead of Non-local for spatial messages (ablation) | spatial graph model · Action Genome SGCls | Recall@20: 50.4 -> 51.1 | Improved |
| insert Non-local layer in the backbone instead of a graph (ablation) | SlowFast R50 · Action Genome SGCls | Recall@20: 48.9 -> 49.1 | No clear effect |
| add temporal connections (tau_c=5, tau_s=5) (ablation) | spatial GAT graph · Action Genome SGCls | Recall@20: 51.1 -> 53.8 | Improved |
| add temporal connections (tau_c=5, tau_s=5) (ablation) | spatial GAT graph · Action Genome PredCls | Recall@10: 79.7 -> 79.3 | No clear effect |
| increase temporal stride tau_s from 2 to 5 (tau_c=3) (ablation) | temporal graph · Action Genome SGCls | Recall@20: 52.9 -> 53.3 | Improved |
| increase temporal context tau_c from 3 to 5 (tau_s=5) (ablation) | temporal graph · Action Genome SGCls | Recall@20: 53.3 -> 53.8 | Improved |
| triple the number of input frames instead of a temporal graph (ablation) | spatial-only graph · Action Genome SGCls | Recall@20: 51.1 -> 52.4 | Improved |
| increase message passing iterations from 1 to 2 (ablation) | spatial GAT graph · Action Genome SGCls | Recall@20: 51.1 -> 51.6 | Improved |
| add message passing among actor nodes (ablation) | SlowFast R50 · AVA v2.2 | mAP: 24.8 -> 25.6 | Improved |
| add implicit context nodes to actor nodes (ablation) | actor-only graph · AVA v2.2 | mAP: 25.6 -> 25.9 | Improved |
| use explicit RPN objects instead of implicit objects (ablation) | context-only graph · AVA v2.2 | mAP: 25.7 -> 25.5 | Hurt |
Closest papers by meaning
- Representation Learning on Visual-Symbolic Graphs for Video Understanding (ECCV 2020 · similarity 0.93)
- Object-Relation Reasoning Graph for Action Recognition (CVPR 2022 · similarity 0.90)
- Action Genome: Actions As Compositions of Spatio-Temporal Scene Graphs (CVPR 2020 · similarity 0.90)
- Temporal Relational Modeling with Self-Supervision for Action Segmentation (AAAI 2021 · similarity 0.89)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.88)
- LASER: A Neuro-Symbolic Framework for Learning Spatio-Temporal Scene Graphs with Weak Supervision (ICLR 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.
