Research map / Label-efficient and robust learning / Noisy labels
Correlated Uncertainty for Learning Dense Correspondences from Noisy Labels
Natalia Neverova, David Novotny, Andrea Vedaldi
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 |
|---|---|---|---|
| train with uncertainty (full model) instead of MSE regression loss | HG 1 stack DensePose · DensePose-COCO 5cm | % points within threshold: 49.92 -> 52.40 | Improved |
| uncertainty training on top of 8-stack hourglass | HG 8 stacks · DensePose-COCO 5cm | % points within threshold: 56.04 -> 56.48 | Improved |
| uncertainty training on 2-stack hourglass at 20cm | HG 2 stacks · DensePose-COCO 20cm | % points within threshold: 85.99 -> 85.58 | No clear effect |
| add elementary uncertainty (simple) versus MSE (ablation) | HG 1 stack · DensePose-COCO UV only (GT parsing) 1cm | % points within threshold: 5.57 -> 6.71 | Improved |
| richer uncertainty models (iid, full) instead of simple/simple-2D (ablation) | HG 1 stack · DensePose-COCO UV accuracy | % points within threshold: 52.67 (simple-2D) -> 52.40 (full) at 5cm | No clear effect |
| richer correlated uncertainty (full) instead of simple (ablation) | uncertainty model · DensePose-COCO human annotations | negative log-likelihood: 1.3159 -> 2.14057 | Improved |
| feed ground truth labels to uncertainty head during training (ablation) | full model · DensePose-COCO UV only (GT parsing) | % points within threshold: 6.78 -> 7.18 (1cm); overall 10cm 74.69 -> 74.28 | Improved |
| fuse ensemble using predicted uncertainty instead of averaging (ablation) | ensemble of simple models · DensePose-COCO 5cm | % points within threshold: 54.26 -> 54.46 | Improved |
Closest papers by meaning
- Correlated Input-Dependent Label Noise in Large-Scale Image Classification (CVPR 2021 · similarity 0.87)
- Training-Free Uncertainty Estimation for Dense Regression: Sensitivity as a Surrogate (AAAI 2022 · similarity 0.87)
- Learning Accurate Dense Correspondences and When To Trust Them (CVPR 2021 · similarity 0.85)
- Uncertainty Sets for Image Classifiers using Conformal Prediction (ICLR 2021 · similarity 0.85)
- An Assessment of Human vs. Model Uncertainty in Soft-Label Learning and Calibration (NeurIPS 2026 · similarity 0.84)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 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.
