Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Metric Learning for Uncertainty Quantification in Image Retrieval
Frederik Rahbæk Warburg, Marco Miani, Silas Brack, Søren Hauberg
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 |
|---|---|---|---|
| online LA instead of post-hoc LA (ablation) | LAM · CIFAR10 (ID) vs SVHN (OoD) | AUROC: 0.50 -> 0.78 | Improved |
| online LA instead of post-hoc LA (ablation) | LAM · FashionMNIST (ID) vs MNIST (OoD) | AUROC: 0.96 -> 0.98 | Improved |
| online LA instead of post-hoc LA (ablation) | LAM · CUB200 (ID) vs CARS196 (OoD) | AUROC: 0.56 -> 0.80 | Improved |
| online LA instead of post-hoc LA (ablation) | LAM · CUB200 (ID) vs CARS196 (OoD) | mAP@5: 0.52 -> 0.48 | Hurt |
| online LA instead of post-hoc LA (ablation) | LAM · LFW (ID) vs CUB200 (OoD) | AUROC: 0.65 -> 0.71 | Improved |
| use Bayesian LA instead of amortized variance (PFE) | LAM (online) · FashionMNIST (ID) vs MNIST (OoD) | AUROC: 0.53 -> 0.98 | Improved |
| use Bayesian LA instead of amortized variance (PFE) | LAM (online) · LFW (ID) vs CUB200 (OoD) | AUROC: 0.03 -> 0.71 | Improved |
| LAM online instead of MC dropout | LAM (online) · CIFAR10 (ID) vs SVHN (OoD) | AUROC: 0.60 -> 0.78 | Improved |
| LAM online instead of MC dropout | LAM (online) · FashionMNIST (ID) vs MNIST (OoD) | ECE: 0.03 -> 0.02 | Improved |
| Arccos distance with fixed Hessian instead of Euclidean fixed (online) (ablation) | LAM · LFW/CUB200 Hessian ablation, ResNet50 2048-d | AUROC: 0.77 -> 0.78 | No clear effect |
| fixed Hessian instead of positive-only Hessian (Arccos, online) (ablation) | LAM · LFW/CUB200 Hessian ablation, ResNet50 2048-d | AUROC: 0.23 -> 0.78 | Improved |
| full Hessian instead of fixed (Arccos, online) (ablation) | LAM · LFW/CUB200 Hessian ablation, ResNet50 2048-d | AUROC: 0.78 -> 0.70 | Hurt |
Closest papers by meaning
- Bayesian Triplet Loss: Uncertainty Quantification in Image Retrieval (ICCV 2021 · similarity 0.88)
- Large-Scale Distance Metric Learning With Uncertainty (CVPR 2018 · similarity 0.86)
- Reliability-Aware Prediction via Uncertainty Learning for Person Image Retrieval (ECCV 2022 · similarity 0.86)
- Introspective Deep Metric Learning (TPAMI 2023 · similarity 0.85)
- Deep Metric Learning with Adaptively Composite Dynamic Constraints (TPAMI 2023 · similarity 0.85)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 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.
