Research map / Label-efficient and robust learning / Self-supervised and contrastive
PUMP: Pyramidal and Uniqueness Matching Priors for Unsupervised Learning of Local Descriptors
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 unsupervised matching-prior training pairs to self-supervised training (ablation) | PUMP descriptors · Aachen Day-Night | localized queries, 0.25m, 2°: 69.63 -> 73.30 | Helped |
| Add unsupervised matching-prior training pairs to self-supervised training (ablation) | PUMP descriptors · ETH3D | AEPE | Mixed |
| Add unsupervised matching-prior training pairs to self-supervised training (ablation) | PUMP descriptors · HPatches viewpoint pairs | AEPE: 4.19 -> 3.76 | Helped |
| Add unsupervised matching-prior training pairs (ablation) | PUMP descriptors with DeepMatching · HPatches viewpoint pairs without interpolation | PCK@1 | Hurt |
| Add style-transfer pairs to SfM120k training (ablation) | PUMP self-supervised and self-supervised-plus-unsupervised models · HPatches and Aachen Day-Night | MMA@1: 37.46 -> 37.97; 37.83 -> 37.64 | Mixed |
| Replace unsupervised training with full supervision on the Aachen-centered training set (ablation) | PUMP ConvMixer descriptors · HPatches | MMA@1: 37.64 -> 36.38 | Hurt |
| Replace unsupervised training with full supervision on the Aachen-centered training set (ablation) | PUMP ConvMixer descriptors · Aachen Day-Night | localized queries, 0.5m, 5°: 87.43 -> 89.01 | Mixed |
| Use the proposed descriptor with SIFT keypoints instead of the SIFT descriptor | SIFT detector · Aachen Day-Night | localized queries, 0.25m, 2°: 45.55 -> 73.30 | Helped |
| Use the proposed descriptor with R2D2 keypoints instead of the R2D2 descriptor | R2D2 detector · Aachen Day-Night | localized queries, 0.25m, 2°: 72.25 -> 73.30 | Helped |
| Use the proposed descriptor with SuperPoint keypoints instead of the SuperPoint descriptor | SuperPoint detector · Aachen Day-Night | localized queries, 0.25m, 2°: 70.16 -> 74.35 | Helped |
Closest papers by meaning
- Comparative Evaluation of Hand-Crafted and Learned Local Features (CVPR 2017 · similarity 0.87)
- Self-Supervised Learning of Geometrically Stable Features Through Probabilistic Introspection (CVPR 2018 · similarity 0.87)
- HyNet: Learning Local Descriptor with Hybrid Similarity Measure and Triplet Loss (NeurIPS 2020 · similarity 0.87)
- ContextDesc: Local Descriptor Augmentation With Cross-Modality Context (CVPR 2019 · similarity 0.87)
- PPFNet: Global Context Aware Local Features for Robust 3D Point Matching (CVPR 2018 · similarity 0.86)
- Local Features and Visual Words Emerge in Activations (CVPR 2019 · 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.