Research map / Video / Video segmentation and tracking
Representation Learning on Visual-Symbolic Graphs for Video Understanding
Effrosyni Mavroudi, Benjamín Béjar Haro, René Vidal
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 visual context module to actor features (ablation) | ST-MPNN · Charades | mAP (%): 10.7 -> 13.7 | Improved |
| add the semantic context module to the visual module (ablation) | ST-MPNN · Charades | mAP (%): 13.7 -> 15.3 | Improved |
| add long-term temporal modeling (ablation) | ST-MPNN · Charades | mAP (%): 15.3 -> 18.6 | Improved |
| add the global model (ablation) | ST-MPNN · Charades | mAP (%): 18.6 -> 23.4 | Improved |
| add semantic context module only (no visual) (ablation) | ST-MPNN · Charades | mAP (%): 10.7 -> 11.7 | Improved |
| add the symbolic graph to the visual graph (ablation) | ST-MPNN · CAD-120 sub-activity | F1-score (%): 89.6 -> 90.4 | Improved |
| add the symbolic graph to the visual graph (ablation) | ST-MPNN · CAD-120 affordance | F1-score (%): 88.6 -> 89.2 | Improved |
| use attention instead of fixed adjacency matrix (ablation) | ST-MPNN · CAD-120 | F1-score (%) | Improved |
| update nodes with both neighbor nodes and incoming edge attributes (ablation) | ST-MPNN · CAD-120 | F1-score (%) | Improved |
| increase number of ST-MPNN layers beyond 4-5 (ablation) | ST-MPNN · CAD-120 | F1-score (%) | No clear effect |
| replace MHA with visual context module in GVD | GVD · ActivityNet Entities val | F1_loc: 24.1 -> 25.2 | Improved |
| add semantic context module to GVD (MHA) | GVD · ActivityNet Entities test | CIDEr: 45.5 -> 47.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- R-C3D — R-C3D: Region Convolutional 3D Network for Temporal Activity Detection (ICCV 2017) · 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
- Unified Graph Structured Models for Video Understanding (ICCV 2021 · similarity 0.93)
- Videos as Space-Time Region Graphs (ECCV 2018 · similarity 0.90)
- G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020 · similarity 0.89)
- Action Genome: Actions As Compositions of Spatio-Temporal Scene Graphs (CVPR 2020 · similarity 0.88)
- LASER: A Neuro-Symbolic Framework for Learning Spatio-Temporal Scene Graphs with Weak Supervision (ICLR 2025 · similarity 0.88)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.86)
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.
