Research map / Video / Video-language understanding
Object Level Visual Reasoning in Videos
Fabien Baradel, Natalia Neverova, Christian Wolf, Julien Mille, Greg Mori
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 object head and ORN module to activity head (ablation) | Object Relation Network · EPIC ablation | top-1 accuracy (%): 38.33 -> 40.89 | Improved |
| add object head and ORN module to activity head (ablation) | Object Relation Network · VLOG ablation | mAP (%): 35.03 -> 37.49 | Improved |
| add object head and ORN module to activity head (ablation) | Object Relation Network · Something-Something ablation | top-1 accuracy (%): 31.31 -> 32.12 | Improved |
| use semantically defined COCO objects instead of pixel-level objects (ablation) | Object Relation Network · EPIC ablation | top-1 accuracy (%): 38.83 -> 40.89 | Improved |
| use semantically defined COCO objects instead of pixel-level objects (ablation) | Object Relation Network · VLOG ablation | mAP (%): 35.18 -> 37.49 | Improved |
| use semantically defined COCO objects instead of pixel-level objects (ablation) | Object Relation Network · Something-Something ablation | top-1 accuracy (%): 31.43 -> 32.12 | Improved |
| make f_phi recurrent (gated RNN) instead of MLP (ablation) | Object Relation Network · EPIC ablation | top-1 accuracy (%): 39.41 -> 40.89 | Improved |
| make f_phi recurrent (gated RNN) instead of MLP (ablation) | Object Relation Network · VLOG ablation | mAP (%): 36.35 -> 37.49 | Improved |
| use visual, shape and class features instead of visual only (ablation) | Object Relation Network · EPIC ablation | top-1 accuracy (%): 38.92 -> 40.89 | Improved |
| use pairwise cliques instead of unary cliques (ablation) | Object Relation Network · EPIC ablation | top-1 accuracy (%): 40.18 -> 40.89 | Improved |
| use pairwise cliques instead of size-3 cliques (ablation) | Object Relation Network · EPIC ablation | top-1 accuracy (%): 37.67 -> 40.89 | Improved |
| use 3D kernels instead of 2D kernels (ablation) | Object Relation Network · Something-Something val | top-1 accuracy (%): 15.73 -> 31.42 | Improved |
Closest papers by meaning
- What Actions Are Needed for Understanding Human Actions in Videos? (ICCV 2017 · similarity 0.90)
- Attend and Interact: Higher-Order Object Interactions for Video Understanding (CVPR 2018 · similarity 0.89)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.89)
- Interacted Object Grounding in Spatio-Temporal Human-Object Interactions (AAAI 2025 · similarity 0.88)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.88)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (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.
