Research map / 3D vision / Depth and stereo
Recurrent MVSNet for High-Resolution Multi-View Stereo Depth Inference
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 |
|---|---|---|---|
| Replace MVSNet 3D CNN regularization with recurrent GRU regularization | MVSNet · DTU evaluation scans | Overall (mm): 0.462 -> 0.422 | Helped |
| Replace MVSNet 3D CNN regularization with recurrent GRU regularization | MVSNet · DTU | GPU memory: 15.4 GB -> 6.7 GB | Helped |
| Use R-MVSNet instead of MVSNet | MVSNet · Tanks and Temples intermediate | Rank: 4 -> 3 | Helped |
| Upsample network depth maps by two before refinement | R-MVSNet · Tanks and Temples intermediate | f_score: 43.48 -> 48.40 | Helped |
| Upsample network depth maps by two before refinement | R-MVSNet · Tanks and Temples advanced | f_score: 24.91 -> 29.55 | Helped |
| Increase depth samples from D=256 to D=512 | R-MVSNet · DTU evaluation | Overall (mm): 0.422mm -> 0.419mm | Helped |
| Fine-tune the DTU-pretrained model on ETH3D low-res training data | R-MVSNet · ETH3D low-res | performance | No clear effect |
| Replace GRU regularization with 3D CNN regularization (ablation) | R-MVSNet · DTU validation | depth map error | Helped |
| Replace GRU regularization with simple spatial regularization (ablation) | R-MVSNet · DTU validation | depth map error | Hurt |
| Replace GRU regularization with winner-take-all selection (ablation) | R-MVSNet · DTU validation | depth map error | Hurt |
| Replace learned image features and cost metric with ZNCC (ablation) | R-MVSNet · DTU validation | depth map error | Hurt |
| Remove variational depth map refinement (ablation) | R-MVSNet · DTU evaluation | f_score | Hurt |
Closest papers by meaning
- RIAV-MVS: Recurrent-Indexing an Asymmetric Volume for Multi-View Stereo (CVPR 2023 · similarity 0.93)
- Fast-MVSNet: Sparse-to-Dense Multi-View Stereo With Learned Propagation and Gauss-Newton Refinement (CVPR 2020 · similarity 0.92)
- Generalized Binary Search Network for Highly-Efficient Multi-View Stereo (CVPR 2022 · similarity 0.91)
- MVS2D: Efficient Multi-View Stereo via Attention-Driven 2D Convolutions (CVPR 2022 · similarity 0.91)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.90)
- MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 2024 · 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.