Research map / Image generation / Autoregressive and latent generation
Learning a Distance Metric from Relative Comparisons between Quadruplets of Images
Marc T. Law, Nicolas Thome, Matthieu Cord
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 |
|---|---|---|---|
| learn a full-matrix metric DM instead of diagonal Dw (ablation) | quadruplet-wise webpage metric · NYT webpage change detection (unsupervised) | MAP: 88.9 -> 93.1 | Improved |
| learn a full-matrix metric DM instead of diagonal Dw (ablation) | quadruplet-wise webpage metric · CNN webpage change detection (unsupervised) | MAP: 88.6 -> 92.2 | Improved |
| learn a full-matrix metric DM instead of diagonal Dw (ablation) | quadruplet-wise webpage metric · NPR webpage change detection (unsupervised) | MAP: 96.5 -> 96.6 | No clear effect |
| use 1-slack cutting plane (structural) instead of projected subgradient for DM (ablation) | quadruplet-wise webpage metric · NYT webpage change detection | MAP: 93.1 -> 92.2 | Hurt |
| add quadruplet-wise constraints instead of triplet-based constraints (ablation) | webpage metric · NYT webpage change detection | MAP: 86.1 -> 88.9 | Improved |
| add quadruplet-wise constraints instead of triplet-based constraints (ablation) | webpage metric · CNN webpage change detection | MAP: 85.2 -> 88.6 | Improved |
| add quadruplet-wise constraints instead of triplet-based constraints (diagonal metric) (ablation) | webpage metric · BBC webpage change detection | MAP: 86.3 -> 86.1 | No clear effect |
| learn a metric instead of Euclidean distance on GIST (ablation) | webpage metric Dw · NYT webpage change detection | MAP: 74.6 -> 88.9 | Improved |
| increase GIST grid resolution (ablation) | unsupervised change detection · webpage change detection | accuracy: accuracy up to 87% on NPR at 10 x 10 | Improved |
| add structural distances (hyperlinks, image URLs) to visual distance (ablation) | unsupervised change detection · CNN webpage change detection | accuracy (%): 72.9 -> 75.0 | Improved |
| add structural distances to visual distance (ablation) | unsupervised change detection · NPR webpage change detection | accuracy (%): 87.0 -> 86.7 | Hurt |
| add 5 annotated pairs per class to unsupervised training (ablation) | semi-supervised change detection · webpage change detection | accuracy: improves accuracy by 5% | Improved |
Closest papers by meaning
- Metric3D: Towards Zero-shot Metric 3D Prediction from A Single Image (ICCV 2023 · similarity 0.82)
- Deep Metric Learning Beyond Binary Supervision (CVPR 2019 · similarity 0.82)
- Understanding and Simplifying Perceptual Distances (CVPR 2021 · similarity 0.81)
- Learning a Deep Color Difference Metric for Photographic Images (CVPR 2023 · similarity 0.81)
- Deep Metric Learning With Angular Loss (ICCV 2017 · similarity 0.81)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.79)
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.
