Research map / Video / Action recognition
EgoPCA: A New Framework for Egocentric Hand-Object Interaction Understanding
Yue Xu, Yong-Lu Li, Zhemin Huang, Michael Xu Liu, Cewu Lu, Yu-Wing Tai, Chi-Keung Tang
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 |
|---|---|---|---|
| pre-train on balanced One4All-P-50K instead of a random 50K subset of the same sources (ablation) | EgoPCA full model · EPIC-100 (one-for-all) | top-1 accuracy: 34.6 -> 41.8 | Improved |
| add SVSA auxiliary loss (ablation) | EgoPCA full model · Ego4D-AR | top-1 accuracy: 16.8 -> 17.6 | Improved |
| add counterfactual reasoning loss (ablation) | EgoPCA full model · EGTEA | top-1 accuracy: 70.3 -> 70.8 | Improved |
| add linear probing (ablation) | EgoPCA full model · Ego4D-AR | top-1 accuracy: 16.8 -> 17.6 | Improved |
| add 10% samples chosen by unified property selection | EgoPCA finetuning · Ego4D-AR | top-1 accuracy: 17.6 -> 18.5 | Improved |
| replace 20% of EGTEA training data with selected samples (ablation) | EgoPCA finetuning · EGTEA Gaze+ | top-1 accuracy: 70.8 -> 68.4 | Hurt |
| replace 5-10% of samples with selected samples (ablation) | EgoPCA finetuning · Ego4D-AR | top-1 accuracy: 17.6 -> 17.9 (-10% => +10%) | Mixed |
| select added data by camera motion/hand location/object location instead of semantics or blurriness (ablation) | EGTEA Gaze+ split 3 · EGTEA Gaze+ split 3 (+10%) | top-1 accuracy: 70.1 (semantics) -> 71.0 (hand location) | Improved |
| use unified weighted property criterion instead of random addition (ablation) | EGTEA Gaze+ split 3 · EGTEA Gaze+ split 3 (+20%) | top-1 accuracy: 69.6 -> 71.2 | Improved |
| pre-train on One4All-P instead of Kinetics-400 (ActionCLIP) | ActionCLIP · EPIC-100 (one-for-all) | top-1 accuracy: 8.8 -> 33.8 | Improved |
| increase pre-train set from 20K to 50K (ablation) | EgoPCA full model · Ego4D-AR (one-for-all) | top-1 accuracy: 6.6 (20K), 6.0 (30K), 7.2 (50K) | Mixed |
Closest papers by meaning
- Are Synthetic Data Useful for Egocentric Hand-Object Interaction Detection? (ECCV 2024 · similarity 0.90)
- Do Egocentric Video-Language Models Truly Understand Hand-Object Interactions? (ICLR 2025 · similarity 0.89)
- Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning (ICLR 2025 · similarity 0.89)
- H+O: Unified Egocentric Recognition of 3D Hand-Object Poses and Interactions (CVPR 2019 · similarity 0.89)
- HOT3D: Hand and Object Tracking in 3D from Egocentric Multi-View Videos (CVPR 2025 · similarity 0.88)
- EgoChoir: Capturing 3D Human-Object Interaction Regions from Egocentric Views (NeurIPS 2024 · 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.
