Research map / Video / Action recognition
Learning State-Aware Visual Representations from Audible Interactions
Himangi Mittal, Pedro Morgado, Unnat Jain, Abhinav Gupta
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 |
|---|---|---|---|
| sample training clips around audio-detected moments of interaction instead of random times (ablation) | RepLAI · EPIC-Kitchens-100 action recognition | verb top-1 accuracy: 28.71 -> 31.71 | Improved |
| sample training clips around audio-detected moments of interaction instead of random times (ablation) | RepLAI · Ego4D state change classification | accuracy: 62.90 -> 66.30 | Improved |
| sample training clips around audio-detected moments of interaction instead of random times (ablation) | RepLAI · Ego4D long-term action anticipation | verb ED@(Z=20) (lower better): 0.792 -> 0.755 | Improved |
| add the AStC audible state-change loss to AVC (ablation) | RepLAI · EPIC-Kitchens-100 action recognition | verb top-1 accuracy: 29.29 -> 31.71 | Improved |
| add the AStC audible state-change loss to AVC (ablation) | RepLAI · Ego4D state change classification | accuracy: 63.60 -> 66.30 | Improved |
| add the AVC loss to AStC (ablation) | RepLAI · EPIC-Kitchens-100 action recognition | verb top-1 accuracy: 29.92 -> 31.71 | Improved |
| add the AVC loss to AStC (ablation) | RepLAI · Ego4D state change classification | accuracy: 64.00 -> 66.30 | Improved |
| initialize from AVID audio-visual pretraining instead of scratch (ablation) | RepLAI · EPIC-Kitchens-100 action recognition | verb top-1 accuracy: 25.75 -> 31.71 | Improved |
| initialize from AVID audio-visual pretraining instead of scratch (ablation) | RepLAI · Ego4D state change classification | accuracy: 66.20 -> 66.30 | No clear effect |
| further train AVID with RepLAI (MoI + AVC + AStC) | AVID · EPIC-Kitchens-100 action recognition | verb top-1 accuracy: 26.62 -> 31.71 | Improved |
| further train AVID with RepLAI | AVID · Ego4D long-term action anticipation | verb ED@(Z=20) (lower better): 0.811 -> 0.755 | Improved |
| RepLAI vs AVID on tail classes | RepLAI · EPIC-Kitchens-100 tail classes | verb top-1 accuracy: 16.80 -> 20.05 | Improved |
Closest papers by meaning
- SoundingActions: Learning How Actions Sound from Narrated Egocentric Videos (CVPR 2024 · similarity 0.90)
- Self-supervised visual learning from interactions with objects (ECCV 2024 · similarity 0.89)
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.89)
- Test-Time Adaptation for Combating Missing Modalities in Egocentric Videos (ICLR 2025 · similarity 0.89)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.88)
- The Audio-Visual Conversational Graph: From an Egocentric-Exocentric Perspective (CVPR 2024 · similarity 0.88)
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.
