Research map / Detection and segmentation / CNN features
Learning Semantic Correspondence Exploiting an Object-Level Prior
Junghyup Lee, Dohyung Kim, Wonkyung Lee, Jean Ponce, Bumsub Ham
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 flow consistency loss to mask consistency loss (ablation) | SFNet · PF-PASCAL test | PCK: 67.5 -> 78.2 | Improved |
| add mask consistency loss to flow consistency loss (ablation) | SFNet · PF-PASCAL test | PCK: 71.8 -> 78.2 | Improved |
| add smoothness term (ablation) | SFNet · PF-PASCAL test | PCK: 78.2 -> 78.7 | Improved |
| replace soft argmax with kernel soft argmax in training and test (ablation) | SFNet with adaptation layer and multi-level features · PF-PASCAL test | PCK: 76.2 -> 78.7 | Improved |
| add adaptation layer to off-the-shelf features with discrete argmax (ablation) | SFNet · PF-PASCAL test | PCK: 45.8 -> 76.8 | Improved |
| apply plain soft argmax directly to off-the-shelf features (ablation) | SFNet without training · PF-PASCAL test | PCK: 45.8 -> 8.8 | Hurt |
| add multi-level conv5-3 features to single-level conv4-23 with kernel soft argmax (ablation) | SFNet · PF-PASCAL test | PCK: 75.0 -> 78.7 | Improved |
| train with bounding-box masks instead of ground-truth foreground masks (ablation) | SFNet · PF-PASCAL test | PCK: 78.7 -> 77.9 | No clear effect |
| train on PF-PASCAL training split (boxes) instead of PASCAL VOC 2012 (ablation) | SFNet · PF-PASCAL test | PCK: 78.7 -> 77.8 | No clear effect |
| train on larger MS COCO subset (16,624 images) instead of 2,791 VOC images (ablation) | SFNet · PF-PASCAL test | PCK: 78.7 -> 77.1 | Hurt |
| use SFNet instead of NCN for runtime on PF-PASCAL | SFNet · PF-PASCAL test | runtime (ms): 284.2 -> 24.9 | Improved |
| use ResNet-101 instead of ResNet-50 backbone | SFNet · PF-PASCAL test | PCK (alpha_img=0.1): 78.3 -> 81.9 | Improved |
Closest papers by meaning
- SFNet: Learning Object-Aware Semantic Correspondence (CVPR 2019 · similarity 0.98)
- SCNet: Learning Semantic Correspondence (ICCV 2017 · similarity 0.92)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.91)
- Pyramidal Semantic Correspondence Networks (TPAMI 2021 · similarity 0.91)
- Pixel-level Semantic Correspondence through Layout-aware Representation Learning and Multi-scale Matching Integration (CVPR 2024 · similarity 0.89)
- 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.
