Research map / 3D vision / Depth and stereo
Learning a Multi-View Stereo Machine
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 |
|---|---|---|---|
| Use geometrically grounded feature projection and unprojection with recurrent multi-view fusion | V-LSM · ShapeNet test, 4 views | mean voxel IoU: 62.1 -> 78.2 | Helped |
| Increase the number of input views from 1 to 4 | V-LSM · ShapeNet test | mean voxel IoU: 61.5 -> 78.2 | Helped |
| Increase the number of input views from 1 to 4 | 3D-R2N2 w/pose · ShapeNet test | mean voxel IoU: 55.1 -> 62.1 | Helped |
| Inject camera pose through fully connected layers | 3D-R2N2 · ShapeNet test, 4 views | mean voxel IoU: 62.0 -> 62.1 | No clear effect |
| Train and test with random image backgrounds instead of white backgrounds (ablation) | V-LSM · ShapeNet test, 1 view | mean voxel IoU: 61.5 -> 60.5 | Hurt |
| Train and test with random image backgrounds instead of white backgrounds (ablation) | V-LSM · ShapeNet test, 4 views | mean voxel IoU: 78.2 -> 75.6 | Hurt |
| Predict per-view depth maps from a globally reasoned 3D representation rather than plane-sweeping stereo | Depth LSM · ShapeNet test, 10 views | median absolute depth error: 0.051 -> 0.024 | Helped |
| Train with synthetically perturbed camera extrinsics (ablation) | V-LSM · ShapeNet test with noisy camera poses | voxel IoU | Helped |
| Use random image backgrounds rather than only white backgrounds (ablation) | V-LSM · ShapeNet test | voxel IoU | Hurt |
Closest papers by meaning
- DeepMVS: Learning Multi-View Stereopsis (CVPR 2018 · similarity 0.90)
- Differentiable Stereopsis: Meshes From Multiple Views Using Differentiable Rendering (CVPR 2022 · similarity 0.90)
- Semantic Multi-View Stereo: Jointly Estimating Objects and Voxels (CVPR 2017 · similarity 0.89)
- Multi-View Stereo Representation Revist: Region-Aware MVSNet (CVPR 2023 · similarity 0.88)
- MVS2D: Efficient Multi-View Stereo via Attention-Driven 2D Convolutions (CVPR 2022 · similarity 0.88)
- PlaneMVS: 3D Plane Reconstruction From Multi-View Stereo (CVPR 2022 · 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.