Research map / 3D vision / Depth and stereo
Learned Multi-Patch Similarity
Wilfried Hartmann, Silvano Galliani, Michal Havlena, Luc Van Gool, Konrad Schindler
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 |
|---|---|---|---|
| use concatenation instead of averaging of Siamese branch outputs (ablation) | OUR (5-view) · DTU CAN | median accuracy error (mm): 0.477 -> 0.478 | No clear effect |
| use averaging instead of concatenation of Siamese branch outputs (ablation) | OUR (5-view) · DTU FLOWER | median accuracy error (mm): 0.843 -> 0.771 | Improved |
| compute proposed similarity over 5 views directly instead of consensus of 2-view scores (ablation) | OUR · DTU (20 objects) | mean completeness error (mm): 2.964 -> 2.126 | Improved |
| compute proposed similarity over 5 views directly instead of consensus of 2-view scores (ablation) | OUR · DTU (20 objects) | median completeness error (mm): 1.391 -> 0.868 | Improved |
| compute proposed similarity over 5 views directly instead of consensus of 2-view scores (ablation) | OUR · DTU (20 objects) | mean accuracy error (mm): 1.339 -> 1.356 | No clear effect |
| replace ZNCC with learned multi-view similarity | plane-sweep stereo · DTU (20 objects) | median accuracy error (mm): 0.577 -> 0.472 | Improved |
| replace ZNCC with learned multi-view similarity | plane-sweep stereo · DTU (20 objects) | mean accuracy error (mm): 1.237 -> 1.356 | Hurt |
| replace LIFT descriptor distance with learned multi-view similarity | plane-sweep stereo · DTU FLOWER | median accuracy error (mm): 1.420 -> 0.771 | Improved |
| replace LIFT descriptor distance with learned multi-view similarity | plane-sweep stereo · DTU CAN | median accuracy error (mm): 1.207 -> 0.478 | Improved |
| increase number of input views from five to nine at test time without retraining (ablation) | OUR · DTU | depth map quality | Improved |
| decrease number of input views from five to three at test time without retraining (ablation) | OUR · DTU | depth map quality | Hurt |
Closest papers by meaning
- PatchmatchNet: Learned Multi-View Patchmatch Stereo (CVPR 2021 · similarity 0.88)
- Patch-Wise Attention Network for Monocular Depth Estimation (AAAI 2021 · similarity 0.88)
- PatchMatch-RL: Deep MVS With Pixelwise Depth, Normal, and Visibility (ICCV 2021 · similarity 0.88)
- Convolutional Patch Representations for Image Retrieval: An Unsupervised Approach (IJCV 2016 · similarity 0.87)
- DVP-MVS: Synergize Depth-Edge and Visibility Prior for Multi-View Stereo (AAAI 2025 · similarity 0.86)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.86)
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.
