Research map / Point clouds / Point cloud classification
Multiview Stereo with Cascaded Epipolar RAFT
Zeyu Ma, Zachary Teed, Jia Deng
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 multiresolution depth fusion (t=0.02) instead of native input only (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 64.38 -> 68.49 | Improved |
| add multiresolution depth fusion (t=0.02) instead of 2x native input only (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 68.08 -> 68.49 | Improved |
| use threshold fusion instead of simple weighted average (w=0.5) (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 66.55 -> 68.49 | Improved |
| use mean instead of max aggregation of neighbor cost volumes (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 57.77 -> 67.36 | Improved |
| use mean instead of std aggregation of neighbor cost volumes (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 59.90 -> 67.36 | Improved |
| use mean instead of std + mean aggregation (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 66.85 -> 67.36 | Improved |
| use adaptive thresholding (p=25%) instead of fixed threshold k=2 (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 66.33 -> 67.36 | Improved |
| use proposed dynamic loss instead of L1 disparity loss (ablation) | CER-MVS · Tanks-and-Temples validation | F1-score: 66.79 -> 67.36 | Improved |
| use proposed dynamic loss instead of direct average of the two losses (ablation) | CER-MVS · Tanks-and-Temples validation | F1-score: 67.32 -> 67.36 | No clear effect |
| use truncated L1 depth loss alone (ablation) | CER-MVS · Tanks-and-Temples validation | F1-score | Hurt |
| increase neighbor views at native resolution from 5 to 15 (25 views at 2x) (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 62.73 -> 67.36 | Improved |
| increase 2x-resolution neighbor views from 5 to 25 (15 views at native) (ablation) | CER-MVS · Tanks-and-Temples validation | Mean F1 (%): 62.73 -> 67.36 | Improved |
Closest papers by meaning
- When Epipolar Constraint Meets Non-Local Operators in Multi-View Stereo (ICCV 2023 · similarity 0.87)
- SD-MVS: Segmentation-Driven Deformation Multi-View Stereo with Spherical Refinement and EM Optimization (AAAI 2024 · similarity 0.85)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.85)
- MUSt3R: Multi-view Network for Stereo 3D Reconstruction (CVPR 2025 · similarity 0.85)
- A Global Depth-Range-Free Multi-View Stereo Transformer Network with Pose Embedding (NeurIPS 2024 · similarity 0.85)
- MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 2024 · 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.
