Research map / 3D vision / Face and shape
EgoGVAE: Ego-body Mesh Reconstruction via Guided Variational Autoencoder
Jaehun Jung, Wonjun Kim
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 latent-distribution guidance from the motion-to-motion VAE (ablation) | EgoGVAE head-to-motion network · AMASS (128 frames) | MPJPE: 128.3 -> 113.7 in MPJPE and 110.3 -> 94.5 in PA-MPJPE | Improved |
| add learnable tokens for unobserved body parts (ablation) | EgoGVAE (with guidance) · AMASS (128 frames) | MPJPE: 113.7 -> 106.7 | Improved |
| add learnable tokens without guidance (ablation) | EgoGVAE baseline · AMASS (128 frames) | MPJPE: 128.3 -> 125.6 | Improved |
| add L_normal_KL regularization toward standard normal (ablation) | EgoGVAE · AMASS (128 frames) | MPJPE: 114.6 -> 106.7 | Improved |
| add velocity loss L_vel (ablation) | EgoGVAE · AMASS (128 frames) | MPJPE: 109.8 -> 106.7 | Improved |
| align latent distributions via joint training instead of learning latent-to-latent mapping (ablation) | EgoGVAE · AMASS (128 frames) | MPJPE: 128.1 -> 106.7 | Improved |
| jointly train guidance network instead of using fixed pre-trained motion prior (ablation) | EgoGVAE · AMASS (128 frames) | MPJPE: 111.2 -> 106.7 | Improved |
| replace diffusion sampling (EgoAllo) with one-step VAE sampling | EgoGVAE · Efficiency (128 frames) | inference time: 1.510 -> 0.026 seconds | Improved |
| EgoGVAE vs EgoAllo on AMASS head-only | EgoGVAE · AMASS (128 frames) | MPJPE: 119.7 -> 106.7 | Improved |
| EgoGVAE vs EgoAllo on RICH cross-dataset | EgoGVAE · RICH (trained on AMASS) | MPJPE: 193.1 -> 179.9 | Improved |
| EgoGVAE vs EgoAllo on real-world EgoBody | EgoGVAE · EgoBody | MPJPE: 172.9 -> 162.9 | Improved |
| EgoGVAE vs EgoEgo jitter on EgoBody | EgoGVAE · EgoBody | Jitter: 1.40 vs 1.42 | 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.
- AvatarPoser — AvatarPoser: Articulated Full-Body Pose Tracking from Sparse Motion Sensing (ECCV 2022) · better on 1 of 1 shared result
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
- Head2Body: Body Pose Generation from Multi-sensory Head-mounted Inputs (ICCV 2025 · similarity 0.88)
- Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data (NeurIPS 2024 · similarity 0.87)
- Stratified Avatar Generation from Sparse Observations (CVPR 2024 · similarity 0.86)
- EgoSim: An Egocentric Multi-view Simulator and Real Dataset for Body-worn Cameras during Motion and Activity (NeurIPS 2024 · similarity 0.85)
- GenHMR: Generative Human Mesh Recovery (AAAI 2025 · similarity 0.85)
- Synthetic Prior for Few-Shot Drivable Head Avatar Inversion (CVPR 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.
