Research map / Detection and segmentation / Open-set and few-shot detection
Leveraging Synthetic Data for Enhancing 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 unsupervised domain adaptation with unlabeled real data to synthetic training | VISOR HOS · VISOR val | Overall AP (Hand+Object): 09.88 -> 33.33 | Improved |
| add synthetic data with semi-supervised domain adaptation (SSDA) to 10% real labels | VISOR HOS · VISOR val (10% real labels) | Overall AP (Hand+Object): 38.55 -> 44.22 | Improved |
| add SSDA to 25% real labels | VISOR HOS · VISOR val (25% real labels) | Overall AP (Hand+Object): 37.90 -> 45.55 | Improved |
| add SSDA to 50% real labels | VISOR HOS · VISOR val (50% real labels) | Overall AP (Hand+Object): 38.15 -> 46.47 | Improved |
| add synthetic pretraining without domain adaptation to 10% real labels | VISOR HOS · VISOR val (10% real labels) | Overall AP (Hand+Object): 38.55 -> 37.62 | Hurt |
| add synthetic pretraining without adaptation, 100% real labels | VISOR HOS · VISOR val (100% real labels) | Overall AP (Hand+Object): 45.33 -> 44.52 | Hurt |
| add FSDA with synthetic data to full real data | VISOR HOS · VISOR val (100% real labels) | Overall AP (Hand+Object): 45.33 -> 46.48 | Improved |
| use ConvNeXt-S instead of ResNet101 backbone | FSDA · VISOR val (100% real labels) | Overall AP (Hand+Object): 46.48 -> 48.36 | Improved |
| use Adaptive Teacher instead of Mean Teacher for SSDA | SSDA · VISOR val (25% real labels) | Overall AP (Hand+Object): 43.69 -> 45.55 | Improved |
| align synthetic objects to target objects (ablation) | UDA (~5k images) · VISOR val | Overall AP (Hand+Object): 31.52 -> 32.60 | Improved |
| align synthetic grasps to target grasps (ablation) | UDA (~5k images) · VISOR val | Overall AP (Hand+Object): 31.52 -> 31.84 | Improved |
| align synthetic environments to target environments (ablation) | UDA (~5k images) · VISOR val | Overall AP (Hand+Object): 31.52 -> 31.59 | No clear effect |
Closest papers by meaning
- Are Synthetic Data Useful for Egocentric Hand-Object Interaction Detection? (ECCV 2024 · similarity 0.98)
- Analyzing the Synthetic-to-Real Domain Gap in 3D Hand Pose Estimation (CVPR 2025 · similarity 0.90)
- Benchmarks and Challenges in Pose Estimation for Egocentric Hand Interactions with Objects (ECCV 2024 · similarity 0.89)
- MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation (NeurIPS 2025 · similarity 0.89)
- EasyHOI: Unleashing the Power of Large Models for Reconstructing Hand-Object Interactions in the Wild (CVPR 2025 · similarity 0.87)
- HOIDiffusion: Generating Realistic 3D Hand-Object Interaction Data (CVPR 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.
