Research map / Label-efficient and robust learning / Self-supervised and contrastive
Neural-Guided RANSAC: Learning Where to Sample Model Hypotheses
Eric Brachmann, Carsten Rother
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 neural guidance to RANSAC sampling | RANSAC · Essential matrix, outdoor and indoor (no side information) | AUC of pose error | Improved |
| train with pose-error task loss instead of hybrid classification-regression loss | NG-RANSAC vs InClass · Essential matrix, outdoor and indoor | AUC of pose error | Improved |
| apply Lowe ratio filter (0.8) to correspondences | RANSAC, USAC and learned methods · Essential matrix, outdoor and indoor | AUC of pose error | Improved |
| add distance ratios as extra network input (+SI) with RootSIFT (ablation) | NG-RANSAC · Essential matrix, outdoor and indoor | AUC of pose error | Improved |
| pass NG-RANSAC learned weights to USAC instead of matching ratios (ablation) | USAC · Essential matrix, outdoor, no ratio filter, M=10^3 | AUC (5/10/20 deg): -0.27/-0.24/-0.34 | Hurt |
| train self-supervised with inlier count instead of ground-truth pose loss (ablation) | NG-RANSAC · Essential matrix, outdoor and indoor | AUC of pose error | Hurt |
| increase hypothesis count from M=10 to M=10^3 (ablation) | NG-RANSAC · Essential matrix runtime | runtime per image pair: 21ms -> 90ms | Hurt |
| choice of training objective (%Inliers, F-score, Mean error) (ablation) | NG-RANSAC · Kitti fundamental matrix | inlier %, F-score, epipolar error | No clear effect |
| add neural guidance to DSAC | DSAC · HLW horizon line estimation | AUC | Improved |
| add neural guidance to RANSAC for fundamental matrix | RANSAC and USAC · Kitti fundamental matrix | inlier %, F-score | Improved |
Closest papers by meaning
- Latent RANSAC (CVPR 2018 · similarity 0.89)
- DSAC - Differentiable RANSAC for Camera Localization (CVPR 2017 · similarity 0.88)
- BANSAC: A Dynamic BAyesian Network for Adaptive SAmple Consensus (ICCV 2023 · similarity 0.87)
- Generalized Differentiable RANSAC (ICCV 2023 · similarity 0.86)
- RANSAC Scoring Done Right (NeurIPS 2026 · similarity 0.85)
- PARSAC: Accelerating Robust Multi-Model Fitting with Parallel Sample Consensus (AAAI 2024 · similarity 0.83)
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.
