Research map / 3D vision / 3D generation
Decomposed Vector-Quantized Variational Autoencoder for Human Grasp Generation
zhao zhe, Mengshi Qi, Huadong Ma
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 |
|---|---|---|---|
| decompose the hand into six parts with separate codebooks (ablation) | VQ-VAE · HO-3D | Quality Index: 7.05 -> 6.76 | Improved |
| add dual-stage decoding (posture then position) to DVQ-VAE (ablation) | DVQ-VAE · HO-3D | Quality Index: 6.76 -> 3.54 | Improved |
| decode position before posture instead of posture first (ablation) | DVQ-VAE+Dual-Stage · HO-3D | Quality Index: 3.54 -> 4.32 | Hurt |
| use two object encoders (type and pose) instead of one (ablation) | DVQ-VAE+Dual-Stage · HO-3D | Quality Index: 6.57 -> 3.54 | Improved |
| use VQ-VAE 2.0 global-feature backbone instead of single-level VQ-VAE (ablation) | VQ-VAE · HO-3D | Quality Index: 7.05 -> 8.46 | Hurt |
| add dual-stage decoding to plain VQ-VAE (two encoders) (ablation) | VQ-VAE · HO-3D | Quality Index: 7.05 -> 3.86 | Improved |
| split hand into six components rather than other part counts (ablation) | DVQ-VAE · HO-3D | multiple metrics | Improved |
| use DVQ-VAE instead of GraspTTA (no test-time optimization) | grasp generation · HO-3D | inference time (s): 19.67 -> 0.14 | Improved |
| DVQ-VAE versus best baseline on Obman quality index | DVQ-VAE · Obman | Quality Index: 2.57 -> 3.07 | Hurt |
| mask part of the object point cloud at test time | DVQ-VAE · HO-3D | Penetration volume (cm^3): 5.66 -> 7.37 | Mixed |
Closest papers by meaning
- 6-DOF GraspNet: Variational Grasp Generation for Object Manipulation (ICCV 2019 · similarity 0.89)
- LatentHOI: On the Generalizable Hand Object Motion Generation with Latent Hand Diffusion. (CVPR 2025 · similarity 0.88)
- MaskDexGrasp: Generative Masked Modeling for Part-Aware Dexterous Grasp Synthesis (CVPR 2026 · similarity 0.88)
- Grasp2Grasp: Vision-Based Dexterous Grasp Translation via Schrödinger Bridges (NeurIPS 2025 · similarity 0.87)
- G-HOP: Generative Hand-Object Prior for Interaction Reconstruction and Grasp Synthesis (CVPR 2024 · similarity 0.87)
- RAGG: Retrieval-Augmented Grasp Generation Model (AAAI 2025 · 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.
