Research map / Detection and segmentation / Open-set and few-shot detection
Are Synthetic Data Useful for Egocentric Hand-Object Interaction Detection?
Rosario Leonardi, Antonino Furnari, Francesco Ragusa, Giovanni Maria Farinella
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 unlabeled real data via unsupervised domain adaptation to synthetic-only training | VISOR HOS · EPIC-KITCHENS VISOR val | Overall Mask AP: 09.88 -> 33.33 | Improved |
| add synthetic data and semi-supervised DA with 10% real labels | VISOR HOS · EPIC-KITCHENS VISOR val (10% labels) | Overall Mask AP: 38.55 -> 44.22 | Improved |
| naively pre-train on synthetic data then fine-tune on 10% real data (no adaptation) (ablation) | VISOR HOS · EPIC-KITCHENS VISOR val (10% labels) | Overall Mask AP: 38.55 -> 37.62 | Hurt |
| add synthetic data with fully supervised domain adaptation | VISOR HOS · EPIC-KITCHENS VISOR val (100% labels) | Overall Mask AP: 45.33 -> 46.48 | Improved |
| add synthetic data with FSDA | VISOR HOS · EgoHOS test (100% labels) | Overall Mask AP: 36.16 -> 39.61 | Improved |
| add synthetic data with SSDA and 10% real labels | VISOR HOS · ENIGMA-51 test (10% labels) | Overall Mask AP: 45.39 -> 57.08 | Improved |
| use in-domain synthetic data instead of out-domain synthetic data (unsupervised) (ablation) | UDA on ENIGMA-51 · ENIGMA-51 test (0% labels) | Overall Mask AP: 6.87 -> 34.78 | Improved |
| use in-domain instead of out-domain synthetic data with some real labels (ablation) | SSDA on ENIGMA-51 · ENIGMA-51 test (10% labels) | Overall Mask AP: 57.08 -> 56.69 | No clear effect |
| replace Adaptive Teacher with Mean Teacher, MT+GRL or Unbiased Teacher (ablation) | SSDA · VISOR val (25% labels) | Overall Mask AP: 45.55 -> 44.32 | Hurt |
| increase synthetic data amount up to about 22k-30k images (ablation) | UDA on VISOR · VISOR val | Overall Mask AP | Improved |
| use simulated hand-object images instead of pasting objects onto ImageNet backgrounds (ablation) | UDA on VISOR · VISOR (3,500 image subset) | Overall Mask AP: 19.62 -> 30.54 | Improved |
| use synthetic source instead of real VISOR data as UDA source for EgoHOS (ablation) | UDA on EgoHOS · EgoHOS test (EPIC-excluded) | Overall Mask AP: 24.39 -> 29.38 | Improved |
Closest papers by meaning
- Leveraging Synthetic Data for Enhancing Egocentric Hand-Object Interaction Detection (IJCV 2026 · similarity 0.98)
- EgoPCA: A New Framework for Egocentric Hand-Object Interaction Understanding (ICCV 2023 · similarity 0.90)
- Analyzing the Synthetic-to-Real Domain Gap in 3D Hand Pose Estimation (CVPR 2025 · similarity 0.87)
- MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation (NeurIPS 2025 · similarity 0.87)
- Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning (ICLR 2025 · similarity 0.86)
- Learning Dense Hand Contact Estimation from Imbalanced Data (NeurIPS 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.
