Research map / 3D vision / Depth and stereo
Continuous 3D Label Stereo Matching Using Local Expansion Moves
Tatsunori Taniai, Yasuyuki Matsushita, Yoichi Sato, Takeshi Naemura
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 RANSAC plane proposer (ablation) | LE-GF · Vintage (V3), nonocc | bad pixel error rate (%): 13.7 -> 5.65 | Improved |
| add RANSAC plane proposer (ablation) | LE-GF · Teddy (V2), nonocc | bad pixel error rate (%): 4.26 -> 3.98 | Mixed |
| add RANSAC proposer on Middlebury V3 training data (ablation) | LE-GF · Middlebury V3 training | bad 2.0 nonocc error: by one point | Improved |
| add regularization term (ablation) | LE-GF · Vintage (V3), nonocc | bad pixel error rate (%): 22.8 -> 5.65 | Improved |
| add regularization term (ablation) | LE-GF · Teddy (V2), nonocc | bad pixel error rate (%): 5.47 -> 3.98 | Improved |
| use local expansion moves instead of fusion-based optimization (Olsson et al.) | Olsson et al. + 3DPM · Teddy (V2), nonocc | bad pixel error rate (%): 5.21 -> 3.98 | Improved |
| use local expansion moves instead of fusion-based optimization (Olsson et al.) | Olsson et al. + 3DPM · Adirondack (V3), nonocc | bad pixel error rate (%): 2.74 -> 1.19 | Improved |
| use guided-filter cost aggregation (LE-GF) instead of bilateral filtering (LE-BF) | LE-BF · Rocks1 | speedup: about 5.3x | Improved |
| parallelize disjoint local alpha-expansions on four CPU cores instead of one | LE-BF and LE-GF · Rocks1 | speedup: about 3.5x | Improved |
| compute unary costs on GPU | LE-BF · Rocks1 | speedup: about 19x | Improved |
| add GPU+CPUx4 over GPU+CPUx1 | LE-BF · Rocks1 | speedup: about 1.7x | Improved |
| use multiple grid-cell sizes (S, M, L) instead of a single size (ablation) | LE-GF · Reindeer | energy and error convergence | Improved |
Closest papers by meaning
- Neural Markov Random Field for Stereo Matching (CVPR 2024 · similarity 0.86)
- Fast Markov Random Field Optimisation for Topologically Noisy 3D Shape Matching (CVPR 2026 · similarity 0.86)
- PatchMatch Filter: Edge-Aware Filtering Meets Randomized Search for Visual Correspondence (TPAMI 2016 · similarity 0.86)
- Digging Into Uncertainty-Based Pseudo-Label for Robust Stereo Matching (TPAMI 2023 · similarity 0.85)
- S2M2: Scalable Stereo Matching Model for Reliable Depth Estimation (ICCV 2025 · similarity 0.85)
- LoS: Local Structure-Guided Stereo Matching (CVPR 2024 · similarity 0.84)
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.
