Research map / Detection and segmentation / Open-set and few-shot detection
Learning Contrastive Representation for Semantic Correspondence
Taihong Xiao, Sifei Liu, Shalini De Mello, Zhiding Yu, Jan Kautz, Ming–Hsuan Yang
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 |
|---|---|---|---|
| add cross-instance pixel-level contrastive learning on PF-PASCAL pairs (ablation) | IC baseline · PF-PASCAL | PCK@0.05: 44.3 -> 51.0 | Improved |
| finetune image-level model on PF-PASCAL pairs without pixel loss (ablation) | IC baseline · PF-PASCAL | PCK@0.05: 44.3 -> 43.7 | No clear effect |
| add pixel-level contrastive loss on self-augmented ImageNet pairs (ablation) | IC baseline · PF-PASCAL | PCK@0.05: 44.3 -> 38.1 | Hurt |
| update backbone with image-level loss instead of freezing it (ablation) | IC + PC · PF-PASCAL | PCK@0.05: 39.5 -> 51.0 | Improved |
| add self-attention module for augmentation (ablation) | Ours · PF-PASCAL | PCK@0.05: 49.9 -> 51.0 | Improved |
| add regularized Hough matching at test time (ablation) | Ours · PF-PASCAL | PCK@0.05: 45.0 -> 51.0 | Improved |
| add OT at test time (ablation) | Ours · PF-PASCAL | PCK@0.05: 41.0 -> 45.0 | Improved |
| use pixel contrastive loss for beam search instead of ground truth (ablation) | Ours · PF-PASCAL | PCK@0.05: 53.4 -> 51.0 | No clear effect |
| add information entropy regularization (ablation) | Ours · PF-WILLOW | PCK@0.05: 40.10 -> 43.10 | Improved |
| add information entropy regularization (ablation) | Ours · PF-PASCAL | PCK@0.05: 51.00 -> 50.70 | Hurt |
| add differentiable OT and RHM layers in training (ablation) | Ours · PF-PASCAL | PCK@0.05: 51.0 -> 32.4 | Hurt |
| add concentration loss in training (ablation) | Ours · PF-PASCAL | PCK@0.05: 51.0 -> 48.1 | No clear effect |
Closest papers by meaning
- Multi-Scale Matching Networks for Semantic Correspondence (ICCV 2021 · similarity 0.91)
- Learning Semantic Correspondence Exploiting an Object-Level Prior (TPAMI 2020 · similarity 0.90)
- Semantic Correspondence: Unified Benchmarking and a Strong Baseline (TPAMI 2025 · similarity 0.90)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.89)
- Neural Matching Fields: Implicit Representation of Matching Fields for Visual Correspondence (NeurIPS 2022 · similarity 0.89)
- Pixel-level Semantic Correspondence through Layout-aware Representation Learning and Multi-scale Matching Integration (CVPR 2024 · similarity 0.89)
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.
