Research map / Learning theory and algorithms / Sample complexity
VSAC: Efficient and Accurate Estimator for H and F
Maksym Ivashechkin, Daniel Barath, Jiří Matas
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 |
|---|---|---|---|
| remove dependent inliers before the randomness test (ablation) | VSAC failure criterion · Non-matching pairs, H (~10^2 iterations) | detected failures (%): 96% -> 100% | Improved |
| remove dependent inliers before the randomness test (ablation) | VSAC failure criterion · Non-matching pairs, F (~10^2 iterations) | detected failures (%): 85% -> 100% | Improved |
| replace DEGENSAC with DEGENSAC+ | VSAC · PhotoTourism (dominant plane) | avg geometric error (px): 0.64 -> 0.44 | Improved |
| replace DEGENSAC with DEGENSAC+ | VSAC · PhotoTourism (dominant plane) | RANSAC run-time (ms): 54.7 -> 31.6 | Improved |
| replace DEGENSAC with DEGENSAC+ | VSAC · PhotoTourism (dominant plane) | degenerate recovery time (ms): 19.8 -> 1.7 | Improved |
| replace DEGENSAC with DEGENSAC+ | VSAC · Kusvod2 (dominant plane) | avg geometric error (px): 2.34 -> 1.83 | Improved |
| replace DEGENSAC with DEGENSAC+ | VSAC · Kusvod2 (dominant plane) | RANSAC run-time (ms): 19.4 -> 11.6 | Improved |
| run local optimization only for non-random, low-Jaccard new best models (ablation) | VSAC · HPatches | LO runs per problem: 4.41 -> 1.00 | Improved |
| use adaptive SPRT instead of fixed-parameter SPRT (ablation) | VSAC · HPatches and PhotoTourism (480 pairs, inlier ratio below 0.5) | geometric error CDF | Improved |
| replace SVD with Gaussian elimination in minimal solver (ablation) | VSAC · H/F minimal model estimation | run time | Improved |
| use IRLS (MAGSAC++) final optimization instead of iterative LSQ | VSAC · HPatches (142 pairs) | median error (px): 0.50 -> 0.43 | Improved |
| use IRLS (MAGSAC++) final optimization instead of iterative LSQ | VSAC · HPatches (142 pairs) | avg run time (ms): 0.9 -> 1.9 | Hurt |
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.
- MAGSAC++ — MAGSAC: Marginalizing Sample Consensus (CVPR 2019) · better on 7 of 8 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
- CONSAC: Robust Multi-Model Fitting by Conditional Sample Consensus (CVPR 2020 · similarity 0.88)
- Latent RANSAC (CVPR 2018 · similarity 0.87)
- PARSAC: Accelerating Robust Multi-Model Fitting with Parallel Sample Consensus (AAAI 2024 · similarity 0.87)
- RANSAC Scoring Done Right (NeurIPS 2026 · similarity 0.85)
- Convex Relaxation for Robust Vanishing Point Estimation in Manhattan World (CVPR 2025 · similarity 0.84)
- Noisy One-point Homographies are Surprisingly Good (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.
