Research map / Video / Action recognition
Multi-Task Learning of Object States and State-Modifying Actions From Web Videos
Tomáš Souček, Jean-Baptiste Alayrac, Antoine Miech, Ivan Laptev, Josef Šivic
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 cross-task negatives (ablation) | N+N multi-classifier · ChangeIt | state precision: 0.36 -> 0.42 | Improved |
| add cross-task negatives (ablation) | N+N multi-classifier · ChangeIt | action precision: 0.68 -> 0.73 | Improved |
| add states as action negatives (ablation) | N+N multi-classifier · ChangeIt | state precision: 0.42 -> 0.41 | No clear effect |
| add actions as state negatives (ablation) | N+N multi-classifier · ChangeIt | action precision: 0.73 -> 0.73 | No clear effect |
| use single joint head without action background class (ablation) | two-head joint classifier · ChangeIt | action precision: 0.67 vs 0.72 | Hurt |
| add states as action negatives (ablation) | two-head joint classifier · ChangeIt | action precision: 0.72 -> 0.75 | Improved |
| add actions as state negatives (ablation) | two-head joint classifier · ChangeIt | state precision: 0.42 -> 0.47 | Improved |
| add actions as state negatives (ablation) | two-head joint classifier · ChangeIt | action precision: 0.75 -> 0.77 | Improved |
| end-to-end finetune ViT-L/14 backbone | Ours · ChangeIt | action precision: 0.75 -> 0.80 | Improved |
| end-to-end finetune ViT-L/14 backbone | Ours · COIN | action precision: 0.77 -> 0.83 | Improved |
| multi-task joint training vs single-task Look for the Change | Ours finetuned · ChangeIt | state precision: 0.35 -> 0.49 | Improved |
| zero-shot transfer from ChangeIt to egocentric | Ours vs Look for the Change · EPIC-KITCHENS | state mAP: 0.12 -> 0.38 | Improved |
Closest papers by meaning
- Look for the Change: Learning Object States and State-Modifying Actions From Untrimmed Web Videos (CVPR 2022 · similarity 0.94)
- Joint Discovery of Object States and Manipulation Actions (ICCV 2017 · similarity 0.89)
- Learning Object State Changes in Videos: An Open-World Perspective (CVPR 2024 · similarity 0.89)
- MOSCATO: Predicting Multiple Object State Change Through Actions (ICCV 2025 · similarity 0.89)
- OnPoint: Offline-to-Online Multi-Level Distillation for Point-Supervised Online Temporal Action Localization (ECCV 2026 · similarity 0.88)
- Understanding Multi-Task Activities from Single-Task Videos (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.
