Research map / 3D vision / Depth and stereo
RIAV-MVS: Recurrent-Indexing an Asymmetric Volume for Multi-View Stereo
Read the paperCheck 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 |
|---|---|---|---|
| Add the residual pose network to Ours(base) (ablation) | RIAV-MVS · ScanNet Test-Set | abs-rel: 0.0885 -> 0.0827 | Helped |
| Add asymmetric transformer attention to Ours(+pose) (ablation) | RIAV-MVS · ScanNet Test-Set | abs-rel: 0.0827 -> 0.0734 | Helped |
| Apply attention symmetrically instead of only to the reference view (ablation) | RIAV-MVS · ScanNet Test-Set | abs-rel: 0.0734 -> 0.0761 | Hurt |
| Add asymmetric attention (ablation) | MVSNet · DTU Test-Set | abs-rel: 0.0143 -> 0.0123 | Helped |
| Add the residual pose network (ablation) | MVSNet · DTU Test-Set | abs-rel: 0.0143 -> 0.0151 | Hurt |
| Add the residual pose network (ablation) | MVSNet · ScanNet Test-Set | abs-rel: 0.1032 -> 0.0937 | Helped |
| Use adaptive depth bins rather than inverse-depth bins (ablation) | RIAV-MVS(+pose) · ScanNet | depth accuracy | Helped |
| Use adaptive depth bins rather than inverse-depth bins (ablation) | RIAV-MVS(+pose,atten) · ScanNet | depth accuracy | Hurt |
| Increase GRU iterations from 16 to 96 (ablation) | RIAV-MVS full model · ScanNet Test-Set | abs-rel: 0.1413 -> 0.1392 | Helped |
| Increase GRU iterations beyond 96 (ablation) | RIAV-MVS full model · ScanNet Test-Set | abs-rel: 0.1392 -> 0.1394 | No clear effect |
| Use 5 views instead of 3 (ablation) | RIAV-MVS(+pose,atten) · 7-Scenes (ScanNet zero-shot) | abs-rel: 0.1084 -> 0.1000 | Helped |
| Use 5 views instead of 3 (ablation) | RIAV-MVS(+pose,atten) · DTU Test-Set | abs-rel: 0.0149 -> 0.0119 | Helped |
Closest papers by meaning
- Recurrent MVSNet for High-Resolution Multi-View Stereo Depth Inference (CVPR 2019 · similarity 0.93)
- MVS2D: Efficient Multi-View Stereo via Attention-Driven 2D Convolutions (CVPR 2022 · similarity 0.91)
- Cost Volume Pyramid Based Depth Inference for Multi-View Stereo (CVPR 2020 · similarity 0.90)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.90)
- A Global Depth-Range-Free Multi-View Stereo Transformer Network with Pose Embedding (NeurIPS 2024 · 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.