Research map / Video / Video-language understanding
SkillSight: Efficient First-Person Skill Assessment with Gaze
Chi Hsuan Wu, Ashutosh Kumar, Kristen Grauman
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 gaze distillation from video+gaze teacher to gaze-only model (ablation) | SkillSight-S · Ego-Exo4D overall | accuracy: 37.0 -> 44.4 | Improved |
| add gaze distillation (student) (ablation) | SkillSight-S · Multi-Sense Badminton | accuracy: 42.3 -> 47.0 | Improved |
| add distillation + action token to gaze-only transformer (ablation) | SkillSight-S · Expert-Novice Soccer | accuracy: 66.0 -> 72.6 | Improved |
| add gaze-guided components (gaze attention, crops, gaze trajectory) to video model | SkillSight-T · Ego-Exo4D overall | accuracy: 45.5 -> 50.1 (vs TimeSformer) | Improved |
| add gaze-guided components | SkillSight-T · Ego-Exo4D Bouldering | accuracy: 28.9 vs 28.0 (TimeSformer) | Mixed |
| gaze-only student instead of video model | SkillSight-S · Ego-Exo4D overall | accuracy: 44.4 vs 45.5 (TimeSformer), 9.5 mW vs 697.5 mW | Mixed |
| use gaze+video student replaced by gaze-only vs EgoDistill/EgoTrigger | SkillSight-S · Ego-Exo4D overall | accuracy: 44.4 vs 42.6 (EgoDistill) | Improved |
| add naive gaze-crop or gaze-pooled processing (EgoExoLearn, Beholder) instead of full-context | EgoExoLearn, Beholder · Ego-Exo4D overall | accuracy: 42.3 and 34.1 vs 45.5 (TimeSformer video-only) | Hurt |
| add gaze when slicing vegetables (cooking) | SkillSight · Ego-Exo4D Cooking | accuracy: SkillSight-S 44.1 vs Gaze-only 40.2 below TimeSformer 56.6 | No clear effect |
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.
- EgoDistill — EgoDistill: Egocentric Head Motion Distillation for Efficient Video Understanding (NeurIPS 2023) · better on 3 of 3 shared results
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
- Gazing Into Missteps: Leveraging Eye-Gaze for Unsupervised Mistake Detection in Egocentric Videos of Skilled Human Activities (CVPR 2025 · similarity 0.88)
- Gaze-VLM: Bridging Gaze and VLMs through Attention Regularization for Egocentric Understanding (NeurIPS 2025 · similarity 0.88)
- In the Eye of MLLM: Benchmarking Egocentric Video Intent Understanding with Gaze-Guided Prompting (NeurIPS 2025 · similarity 0.87)
- GazeFlow: From Human Gaze Behavior to Generative Egocentric Gaze Prediction (NeurIPS 2026 · similarity 0.86)
- Sharingan: A Transformer Architecture for Multi-Person Gaze Following (CVPR 2024 · similarity 0.85)
- EyEar: Learning Audio Synchronized Human Gaze Trajectory Based on Physics-Informed Dynamics (AAAI 2025 · similarity 0.85)
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.
