Research map / 3D vision / Depth and stereo
IS-MVSNet: Importance Sampling-Based MVSNet
Likang Wang, Yue Gong, Xinjun Ma, Qirui Wang, Kaixuan Zhou, Lei Chen
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 importance sampling (geometric interval) instead of uniform sampling (ablation) | IS-MVSNet · ETH3D training set (2cm), stage 3 | F-score: 0.44 -> 0.58 | Improved |
| add importance sampling instead of uniform sampling (ablation) | IS-MVSNet · ETH3D training set (2cm), stage 2 | F-score: 0.32 -> 0.41 | Improved |
| add importance sampling instead of uniform sampling (ablation) | IS-MVSNet · ETH3D training set (2cm), stage 1 | F-score: 0.14 -> 0.14 | No clear effect |
| convert i.i.d. uniform sampling to identical interval sampling (ablation) | uniform sampling · ETH3D high-res training set | F-score: 6.4% F-score improvement | Improved |
| sample depths from zero-mean Gaussian (i.i.d.) rather than uniform (ablation) | IS-MVSNet · ETH3D | F-score | Improved |
| increase k above 1 (denser sampling near previous prediction) (ablation) | IS-MVSNet · ETH3D and DTU | F-score / overall distance | Mixed |
| select k with unsupervised ternary search on photometric variance (ablation) | IS-MVSNet · ETH3D | F-score | Improved |
| use two reference views per scene for k estimation (ablation) | IS-MVSNet · ETH3D | variance estimate | No clear effect |
| use importance sampling in place of uniform sampling in Vis-MVSNet | IS-MVSNet · DTU testing set | overall distance (mm): 0.365 -> 0.355 | Improved |
| use importance sampling in place of uniform sampling in Vis-MVSNet | IS-MVSNet · Tanks & Temples intermediate | F-score mean (%): 60.03 -> 62.82 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PatchmatchNet — PatchmatchNet: Learned Multi-View Patchmatch Stereo (CVPR 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Fast-MVSNet: Sparse-to-Dense Multi-View Stereo With Learned Propagation and Gauss-Newton Refinement (CVPR 2020 · similarity 0.89)
- Point-Based Multi-View Stereo Network (ICCV 2019 · similarity 0.88)
- P-MVSNet: Learning Patch-Wise Matching Confidence Aggregation for Multi-View Stereo (ICCV 2019 · similarity 0.87)
- EC-MVSNet: Enhanced Cascaded Multi-View Stereo with Cross-Scale Relevance Integration (AAAI 2026 · similarity 0.87)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.85)
- MVSAnywhere: Zero-Shot Multi-View Stereo (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.
