Research map / Learning theory and algorithms / Clustering
Superpixel-Based Two-View Deterministic Fitting for Multiple-Structure Data
Guobao Xiao, Hanzi Wang, Yan Yan, David Suter
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 SDF instead of RANSAC on a low-inlier pair | RANSAC · Bonython (homography, single-structure) | Sampson error: 10.76 -> 0.32 | Improved |
| use SDF instead of Astar (deterministic) on low-inlier pairs | Astar · Bonython and Physics (homography) | solved within 1 hour: Astar >3600 s -> SDF 0.36 s on Bonython | Improved |
| use SDF instead of RANSAC on high-inlier pair | RANSAC · Keble (homography, single-structure) | CPU time (s): 0.29 -> 0.42 | Hurt |
| use SDF instead of RANSAC on multi-structure fundamental matrix | RANSAC · Multi-structure datasets D6-D10 | CPU time: about 32.08~52.37 times faster | Improved |
| use SDF instead of T-linkage on multi-structure data | T-linkage · MC3 (D4, homography) | CPU time: 1508.62 s -> 1.96 s | Improved |
| use SDF instead of T-linkage on multi-structure fundamental matrix | T-linkage · Multi-structure datasets D6-D10 | CPU time: 42.48~75.18 times faster | Improved |
| use SDF instead of PROSAC on MC3 | PROSAC · MC3 (D4, homography) | fit success within 1 hour: PROSAC >3600 s -> SDF 1.96 s | Improved |
| increase the outlier percentage above 50% | SDF vs RANSAC/PROSAC · Physics and Book image pairs | CPU time: SDF time roughly unchanged while RANSAC grows above 40% (homography) and 50% (fundamental) | Improved |
| increase number of superpixels M above 300 (ablation) | SDF · Physics (homography) | Sampson error: M larger than 300 raises error dramatically | Hurt |
| vary number of superpixels M within [50, 300] (ablation) | SDF · five of six image pairs | Sampson error: low error across M | No clear effect |
| use SDF instead of T-linkage for accuracy on D1, D2, D5 | T-linkage · Multi-structure homography | Sampson error: 1.03 vs 1.02 (D1); 1.47 vs 1.36 (D5) | Hurt |
Closest papers by meaning
- Superpixel-Guided Two-View Deterministic Geometric Model Fitting (IJCV 2018 · similarity 0.97)
- Extract and Merge: Superpixel Segmentation with Regional Attributes (ECCV 2020 · similarity 0.84)
- Bayesian Adaptive Superpixel Segmentation (ICCV 2019 · similarity 0.84)
- Efficient Deterministic Search with Robust Loss Functions for Geometric Model Fitting (TPAMI 2021 · similarity 0.84)
- Hierarchical Histogram Threshold Segmentation - Auto-terminating High-detail Oversegmentation (CVPR 2024 · similarity 0.83)
- Learning Invariant Inter-pixel Correlations for Superpixel Generation (AAAI 2024 · similarity 0.82)
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.
