Research map / Label-efficient and robust learning / Recognition and long-tail
Demystifying Unsupervised Semantic Correspondence Estimation
Mehmet Aygün, Oisin Mac Aodha
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 |
|---|---|---|---|
| use asymmetric temperatures (tau1<tau2) with distance loss instead of LEAD (ablation) | LEAD, sup. CNN · SPair-71k | PCK@0.1 (%): 31.7 -> 34.0 | Improved |
| use asymmetric temperatures instead of LEAD (ablation) | LEAD, sup. CNN · CUB | PCK@0.1 (%): 51.5 -> 60.8 | Improved |
| use asymmetric temperatures instead of LEAD (ablation) | LEAD, sup. CNN · AFLW | PCK@0.1 (%): 58.0 -> 63.6 | Improved |
| use asymmetric temperatures instead of LEAD (ablation) | LEAD, unsup. Transformer · SPair-71k | PCK@0.1 (%): 33.6 -> 32.9 | Hurt |
| change LEAD temperature (ablation) | LEAD · SPair-71k | PCK@0.1 (%): 31.7, 31.9, 31.4 | No clear effect |
| increase tau1,tau2 from 0.2,0.4 to 0.4,0.8 (ablation) | ASYM · SPair-71k | PCK@0.1 (%): 34.0 -> 29.5 | Hurt |
| increase CL temperature from 0.05 to 0.2 (ablation) | CL · SPair-71k | PCK@0.1 (%): 8.2 -> 26.6 | Improved |
| use MSE instead of cross-entropy for LEAD (tau=0.05) (ablation) | LEAD · SPair-71k | PCK@0.1 (%): 31.7 -> 31.5 | No clear effect |
| use MSE instead of cross-entropy for ASYM (ablation) | ASYM · SPair-71k | PCK@0.1 (%): 32.0 -> 34.0 | Improved |
| use conv3 features instead of conv4 (ablation) | ResNet-50 features, no projection · SPair-71k | PCK@0.1 (%): 15.8 -> 31.8 | Improved |
| use unsupervised pre-trained Transformer instead of supervised (ablation) | no projection · SDogs | PCK@0.1 (%): 38.0 -> 42.7 | Improved |
| penalize swap errors (PCK-dagger instead of PCK) | ASYM · SPair-71k | PCK-dagger vs PCK: 34.0 -> 27.2 | Hurt |
Closest papers by meaning
- Semantic Correspondence: Unified Benchmarking and a Strong Baseline (TPAMI 2025 · similarity 0.91)
- Unsupervised Semantic Correspondence Using Stable Diffusion (NeurIPS 2023 · similarity 0.89)
- It's a (Blind) Match! Towards Vision-Language Correspondence without Parallel Data (CVPR 2025 · similarity 0.89)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 2025 · similarity 0.88)
- Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence (CVPR 2024 · similarity 0.87)
- SD4Match: Learning to Prompt Stable Diffusion Model for Semantic Matching (CVPR 2024 · 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.
