Research map / Vision-language and multimodal / Human and embodied modeling
Semantic Correspondence: Unified Benchmarking and a Strong Baseline
Kaiyan Zhang, Xinghui Li, Jingyi Lu, Kai Han
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 SD2-1 features to DINOv2 features (concatenate) | DINOv2 zero-shot · SPair-71k, 840/60 res | PCK@0.1: 54.4 -> 60.4 | Improved |
| fuse SD2-1 with DINOv2 at low resolution | DINOv2 zero-shot · SPair-71k, 224/16 res | PCK@0.1: 43.1 -> 41.9 | Hurt |
| fine-tune last layer of DINOv2 | DINOv2 baseline · SPair-71k, 840/60 res | PCK@0.1: 54.4 -> 81.7 | Improved |
| add ResNet bottleneck block enhancement (ablation) | frozen DINOv2 · SPair-71k, 840/60 res | PCK@0.1: 54.4 -> 75.2 (+20.8) | Improved |
| add ResNet bottleneck block enhancement (ablation) | fine-tuned DINOv2 · SPair-71k, 840/60 res | PCK@0.1: 81.7 -> 84.2 (+2.5) | Improved |
| add self-attention enhancement (ablation) | fine-tuned DINOv2 · SPair-71k, 840/60 res | PCK@0.1: 81.7 -> 80.9 (-0.8) | Hurt |
| add self-attention enhancement at low resolution (ablation) | frozen DINOv2 · SPair-71k, 224/16 res | PCK@0.1: +1.7 (vs +22.1 at 840, 60) | Mixed |
| add NeighConsensus cost aggregator (ablation) | frozen DINOv2 · SPair-71k, 840/60 res | PCK@0.1: 54.4 -> 61.9 (+7.5) | Improved |
| add NeighConsensus cost aggregator (ablation) | fine-tuned DINOv2 · SPair-71k, 840/60 res | PCK@0.1: 81.7 -> 85.2 (+3.5) | Improved |
| add Match2Match cost aggregator (ablation) | frozen DINOv2 · SPair-71k, 840/60 res | PCK@0.1: 54.4 -> 54.1 (-0.3) | No clear effect |
| add CATs cost aggregator (ablation) | fine-tuned DINOv2 · SPair-71k, 448/32 res | PCK@0.1: 76.6 -> 71.4 (-5.2) | Hurt |
| use original NeighConsensus instead of reduced version (ablation) | fine-tuned DINOv2 + NC · SPair-71k, 224/16 res | PCK@0.1: 62.4 -> 66.1 | Improved |
Closest papers by meaning
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.91)
- SOCO: Benchmarking Semantic Object Correspondence in Vision Foundation Models (ECCV 2026 · similarity 0.91)
- Pixel-level Semantic Correspondence through Layout-aware Representation Learning and Multi-scale Matching Integration (CVPR 2024 · similarity 0.90)
- Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence (CVPR 2024 · similarity 0.89)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 2025 · similarity 0.88)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.88)
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.
