Research map / Video / Video segmentation and tracking
Generative Adversarial Network for Future Hand Segmentation from Egocentric Video
Wenqi Jia, Miao Liu, James M. Rehg
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 GAN-generated future head motion to 3DFCN (ablation) | 3DFCN · EPIC-Kitchens 55 | F1 score: +0.9%/1.5%/1.8% F1 (short/middle/long) | Improved |
| add GAN-generated future head motion to 3DFCN (ablation) | 3DFCN · EGTEA Gaze+ | F1 score: +0.8%/0.9%/0.7% F1 (short/middle/long) | Improved |
| train head motion generator adversarially instead of L1 regression (ablation) | EgoGAN · EPIC-Kitchens 55 | F1 score: 0.7%/1.7%/1.7% F1 lower for HeadReg | Improved |
| train head motion generator adversarially instead of L1 regression (ablation) | EgoGAN · EGTEA Gaze+ | F1 score: 1.1%/1.5%/1.5% F1 lower for HeadReg | Improved |
| apply GAN to head motion rather than directly to hand masks (ablation) | EgoGAN · EGTEA Gaze+ | F1 score: DirectGan slightly decreases 3DFCN on EGTEA, minor improvement on EPIC-Kitchens | Improved |
| model future head motion instead of future gaze (ablation) | EgoGAN · EGTEA Gaze+ | F1 score: 0.7%/0.6%/0.6% F1 lower for future gaze | Improved |
| switch 3DFCN backbone from I3D-Res50 to I3D-Res101 (ablation) | 3DFCN · EPIC-Kitchens 55 | F1 score: +0.1%/0.3%/0.3% F1 | No clear effect |
| switch 3DFCN backbone from I3D-Res50 to I3D-Res101 (ablation) | 3DFCN · EGTEA Gaze+ | F1 score: +0.7%/0.4%/0.5% F1 | Improved |
| add EgoGAN to I3D-Res101 3DFCN (ablation) | 3DFCN I3D-Res101 · EPIC-Kitchens 55 | F1 score: +0.5%/0.4%/0.7% F1 | Improved |
| add EgoGAN to I3D-Res101 3DFCN (ablation) | 3DFCN I3D-Res101 · EGTEA Gaze+ | F1 score: +0.1%/0.1%/0.3% F1 | Improved |
| model hand motion over time instead of zero-motion baseline (ablation) | EgoGAN · EPIC-Kitchens 55 | F1 score: 46.14% and 34.30% F1 for middle and long term, 1.9%/3.8% lower than full model | Improved |
| use GAN rather than regression for head motion generation (ablation) | EgoGAN · EPIC-Kitchens 55 head motion | EPE: 10.39 -> 7.08 | Improved |
Closest papers by meaning
- Deep Future Gaze: Gaze Anticipation on Egocentric Videos Using Adversarial Networks (CVPR 2017 · similarity 0.90)
- Anticipating Where People will Look Using Adversarial Networks (TPAMI 2018 · similarity 0.87)
- Generalizing Hand Segmentation in Egocentric Videos With Uncertainty-Guided Model Adaptation (CVPR 2020 · similarity 0.86)
- Dual Motion GAN for Future-Flow Embedded Video Prediction (ICCV 2017 · similarity 0.86)
- Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data (NeurIPS 2024 · similarity 0.85)
- EgoExo-Gen: Ego-centric Video Prediction by Watching Exo-centric Videos (ICLR 2025 · similarity 0.84)
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.
