Research map / Learning theory and algorithms / Prediction and decision losses
Quantitative Evaluation of Confidence Measures in a Machine Learning World
Matteo Poggi, Fabio Tosi, Stefano Mattoccia
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 learning-based CCNN instead of best conventional measure APKR11 | confidence measures with AD-CENSUS · K12 | AUC: 0.1806 -> 0.1223 | Improved |
| use learning-based CCNN instead of APKR11 | confidence measures with MC-CNN · K12 | AUC: 0.0566 -> 0.0297 | Improved |
| train ENS in regression mode (ENSr) instead of classification (ENSc) | ensemble learning confidence · K12 (AD-CENSUS) | AUC rank: 7 -> 5 | Improved |
| use ENSc rather than ENSr in cost modulation | SGM cost modulation · K12/K15/M14 | bad3 / bad1 error: 10.42 -> 10.635 | Improved |
| add confidence-based cost modulation using CCNN to SGM | SGM · K12 | bad3 error rate: 16.53 -> 10.613 | Improved |
| add confidence-based cost modulation using CCNN to SGM | SGM · M14 | bad1 error rate: 25.91 -> 22.861 | Improved |
| add confidence-based cost modulation using CCNN to SGM | SGM · K15 | bad3 error rate: 13.68 -> 8.791 | Improved |
| increase training samples from 5 to 25 K12 stereo pairs for random-forest measures (ablation) | ENSr, GCP, LEV, O1 · K12 (AD-CENSUS) | Delta_k: less than 0.05 improvement | No clear effect |
| increase CNN-based measure training data to 2.7 million samples (ablation) | CCNN, PBCP · K12 (AD-CENSUS) | Delta_k: 0.7 -> 2.7 million samples | Improved |
| use small training subset for CCNN (ablation) | CCNN · K12 (AD-CENSUS) | Delta_k: about 0.25 margin | Hurt |
| train on M14 instead of K12 for generalization test (ablation) | ENSr, GCP, LEV vs CCNN, PBCP, O1 · M14 (AD-CENSUS) | Delta_k | Mixed |
| use full-patch size variants versus naive (second minimum) variants | conventional confidence measures · K12/K15/M14 | AUC | Hurt |
Closest papers by meaning
- Learning to Predict Stereo Reliability Enforcing Local Consistency of Confidence Maps (CVPR 2017 · similarity 0.95)
- Using Self-Contradiction to Learn Confidence Measures in Stereo Vision (CVPR 2016 · similarity 0.92)
- On the confidence of stereo matching in a deep-learning era: a quantitative evaluation (TPAMI 2021 · similarity 0.91)
- Learning and Selecting Confidence Measures for Robust Stereo Matching (TPAMI 2018 · similarity 0.91)
- Modeling Stereo-Confidence out of the End-to-End Stereo-Matching Network via Disparity Plane Sweep (AAAI 2024 · similarity 0.90)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.86)
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.
