Research map / 3D vision / Depth and stereo
Fast-MVSNet: Sparse-to-Dense Multi-View Stereo With Learned Propagation and Gauss-Newton Refinement
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 sparse high-resolution depth instead of low-resolution depth (ablation) | Fast-MVSNet · DTU evaluation | Overall (mm): 0.537 -> 0.436 | Helped |
| add learned depth propagation (ablation) | Fast-MVSNet sparse high-resolution depth · DTU evaluation | Overall (mm): 0.436 -> 0.409 | Helped |
| add Gauss-Newton refinement (ablation) | Fast-MVSNet with propagation · DTU evaluation | Overall (mm): 0.409 -> 0.370; 9.5% | Helped |
| replace low-resolution depth and PointFlow with sparse high-resolution depth, learned propagation, and Gauss-Newton (ablation) | Point-MVSNet-style comparison · DTU evaluation | Overall (mm): 0.391 -> 0.370 | Helped |
| use one Gauss-Newton iteration instead of none (ablation) | Fast-MVSNet · DTU evaluation | Overall (mm): 0.4077 -> 0.3695 | Helped |
| increase Gauss-Newton iterations beyond one (ablation) | Fast-MVSNet · DTU evaluation | Overall (mm): 0.3695 -> 0.3659 | Helped |
| increase the number of Gauss-Newton iterations (ablation) | Fast-MVSNet · DTU evaluation | Acc. (mm): 0.3360 -> 0.3443 | Hurt |
| increase the geometric consistency threshold η (ablation) | Fast-MVSNet depth-map fusion · DTU evaluation | Acc. (mm): 0.3360 -> 0.8381 | Hurt |
| increase the geometric consistency threshold η (ablation) | Fast-MVSNet depth-map fusion · DTU evaluation | Comp. (mm): 0.4030 -> 0.3187 | Helped |
| require visibility in more views (ablation) | Fast-MVSNet depth-map fusion · DTU evaluation | Acc. (mm): 0.3360 -> 0.3007 | Helped |
| require visibility in more views (ablation) | Fast-MVSNet depth-map fusion · DTU evaluation | Comp. (mm): 0.4030 -> 0.5212 | Hurt |
| use Fast-MVSNet instead of Point-MVSNet | DTU evaluation · DTU evaluation | Runtime (s): 3.35 -> 0.6 | Helped |
Closest papers by meaning
- MVS2D: Efficient Multi-View Stereo via Attention-Driven 2D Convolutions (CVPR 2022 · similarity 0.93)
- Generalized Binary Search Network for Highly-Efficient Multi-View Stereo (CVPR 2022 · similarity 0.93)
- Recurrent MVSNet for High-Resolution Multi-View Stereo Depth Inference (CVPR 2019 · similarity 0.92)
- MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 2024 · similarity 0.89)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.89)
- MVSAnywhere: Zero-Shot Multi-View Stereo (CVPR 2025 · similarity 0.88)
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.