Research map / Video / Action recognition
Discovering Novel Actions from Open World Egocentric Videos with Object-Grounded Visual Commonsense Reasoning
Sanjoy Kundu, Shubham Trehan, Sathyanarayanan N Aakur
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 evidence-based CLIP object grounding via ConceptNet instead of KGL prior-only grounding | KGL · GTEA Gaze | Object accuracy: 5.12 -> 13.07 | Improved |
| naively add CLIP grounding to KGL | KGL · GTEA Gaze / GazePlus | Action accuracy: 8.04 -> 8.15 on Gaze; average 2% gain | No clear effect |
| add posterior-based action refinement with visual-semantic grounding | ALGO · GTEA Gaze | Action accuracy: KGL 8.04 -> ALGO 17.05 | Improved |
| add LaViLa action priors to ALGO | ALGO · GTEA Gaze | Activity accuracy: 15.05 -> 22.05 | Improved |
| add LaViLa action priors to ALGO on EPIC-Kitchens | ALGO · EPIC-Kitchens-100 | Object accuracy: verb improved to 12.54; object 22.84 vs LaViLa 23.25 | Mixed |
| run posterior refinement for more than two iterations (ablation) | ALGO · GTEA Gaze | Activity accuracy | No clear effect |
| use ConceptNet Numberbatch instead of GloVe embeddings (ablation) | ALGO action grounding · GTEA Gaze | activity accuracy | Improved |
| use S3D-G instead of S3D visual features (ablation) | ALGO action grounding · GTEA Gaze | activity accuracy | No clear effect |
| use CLIP + evidence grounding instead of CLIP only (ablation) | ALGO object grounding · GTEA Gaze | Object accuracy | Improved |
| replace dataset-derived search space with an unknown GPT-4 generated space | ALGO · GTEA Gaze | Verb accuracy: 14.11 -> 9.87 | Hurt |
Closest papers by meaning
- Opening the Vocabulary of Egocentric Actions (NeurIPS 2023 · similarity 0.89)
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.88)
- EgoThinker: Unveiling Egocentric Reasoning with Spatio-Temporal CoT (NeurIPS 2025 · similarity 0.87)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.87)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 2025 · similarity 0.87)
- Object-Shot Enhanced Grounding Network for Egocentric Video (CVPR 2025 · 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.
