Research map / 3D vision / Human pose estimation
MobRecon: Mobile-Friendly Hand Mesh Reconstruction From Monocular Image
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use DenseStack rather than MobileNet-Stack for 2D encoding (ablation) | MobRecon · FreiHAND | PA-MPJPE (mm): 13.68 -> 7.77 | Helped |
| Pre-train DenseStack with the synthetic complement dataset and fine-tune on FreiHAND+ours (ablation) | MobRecon · FreiHAND | PA-MPJPE (mm): 7.77 -> 7.55 | Helped |
| Use MapReg instead of direct position regression (ablation) | MobRecon feature lifting module · FreiHAND | PA-MPJPE (mm): 7.11 -> 6.95 | Helped |
| Add a skip connection to the heatmap representation (ablation) | MobRecon feature lifting module · FreiHAND | PA-MPJPE (mm): 7.41 -> 7.28 | Helped |
| Use soft-argmax with heatmaps rather than heatmap joint-wise pooling (ablation) | MobRecon feature lifting module · FreiHAND | PA-MPJPE (mm): 7.41 -> 7.11 | Helped |
| Use pose-to-vertex lifting (PVL) and grid sampling rather than global pooling and a fully connected mapping (ablation) | MobRecon feature lifting module · FreiHAND | PA-MPJPE (mm): 7.55 -> 6.95 | Helped |
| Add 3D consistency loss (ablation) | MobRecon · HO3Dv2 | 3D acceleration: 5.41 -> 4.78 | Helped |
| Add 2D consistency loss alongside 3D consistency loss (ablation) | MobRecon · HO3Dv2 | 3D acceleration: 4.78 -> 4.75 | Helped |
| Replace SpiralConv++ with depth-separable spiral convolution (ablation) | MobRecon 3D decoder · FreiHAND | PA-MPJPE (mm): 6.85 -> 6.87 | No clear effect |
| Replace SpiralConv++ with depth-separable spiral convolution (ablation) | MobRecon 3D decoder · Apple A14 CPU | FPS: 59 -> 67 | Helped |
| Use GhostStack instead of DenseStack (ablation) | MobRecon · FreiHAND | PA-MPJPE (mm): 6.9 -> 8.8 | Hurt |
| Use GhostStack with DSConv rather than DenseStack with DSConv (ablation) | MobRecon · Apple A14 CPU | FPS: 67 -> 83 | Helped |
Closest papers by meaning
- Fine-Grained Multi-View Hand Reconstruction Using Inverse Rendering (AAAI 2024 · similarity 0.87)
- A Simple Baseline for Efficient Hand Mesh Reconstruction (CVPR 2024 · similarity 0.87)
- Reconstructing Hands in 3D with Transformers (CVPR 2024 · similarity 0.87)
- HandOS: 3D Hand Reconstruction in One Stage (CVPR 2025 · similarity 0.87)
- HandBooster: Boosting 3D Hand-Mesh Reconstruction by Conditional Synthesis and Sampling of Hand-Object Interactions (CVPR 2024 · similarity 0.86)
- SiMA-Hand: Boosting 3D Hand-Mesh Reconstruction by Single-to-Multi-View Adaptation (AAAI 2024 · 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.