Research map / 3D vision / Depth and stereo
SMVRT: Implicit Human 3D Modeling Using Sparse Multi-View Volumetric Reconstruction with Transformer Fusion
Chuanmao Fan, Chenxi Zhao, Ye Duan
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 MMFT multi-view multi-level 2D-to-3D fusion transformer (ablation) | SMVRT no-fusion baseline · THUman2.1 | Chamfer L1 mean: 4.40 -> 3.84 | Improved |
| add QFT query point fusion transformer (ablation) | SMVRT no-fusion baseline · THUman2.1 | Chamfer L1 mean: 4.40 -> 4.02 | Improved |
| add AFM alternating global-local 2D fusion (ablation) | SMVRT no-fusion baseline · THUman2.1 | Chamfer L1 mean: 4.40 -> 4.37 | No clear effect |
| add AFM on top of MMFT+QFT (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.61 -> 3.58 | No clear effect |
| combine MMFT and QFT (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.84 -> 3.61 | Improved |
| increase input views from 4 to 8 (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.58 -> 2.96 | Improved |
| reduce input views from 4 to 2 (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.58 -> 8.74 | Hurt |
| raise image resolution to 1024 with 8 views (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 2.96 -> 2.91 | No clear effect |
| add Sapiens normal maps as input (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.90 -> 3.71 | Improved |
| add depth, then Plucker rays, to the inputs (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.71 -> 3.65 -> 3.58 | Improved |
| use Mamba instead of Transformer blocks in AFM (ablation) | SMVRT · THUman2.1 | Chamfer L1 mean: 3.577 -> 3.586 | No clear effect |
| add low-level 2D detail features at query points (ablation) | QFT · THUman2.1 | Chamfer L1 mean: 3.92 -> 3.58 | Improved |
Closest papers by meaning
- SVTformer: Spatial-View-Temporal Transformer for Multi-View 3D Human Pose Estimation (AAAI 2025 · similarity 0.87)
- RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images (CVPR 2025 · similarity 0.86)
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.86)
- MagicMan: Generative Novel View Synthesis of Humans with 3D-Aware Diffusion and Iterative Refinement (AAAI 2025 · similarity 0.86)
- FusionFormer: A Concise Unified Feature Fusion Transformer for 3D Pose Estimation (AAAI 2024 · similarity 0.86)
- GoLF-NRT: Integrating Global Context and Local Geometry for Few-Shot View Synthesis (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.
