Research map / Detection and segmentation / Weakly supervised detection and segmentation
PARN: Pyramidal Affine Regression Networks for Dense Semantic Correspondence
Sangryul Jeon, Seungryong Kim, Dongbo Min, Kwanghoon Sohn
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 level 2 on top of level 1 (ablation) | PARN (VGG-16) · Taniai benchmark (average) | matching accuracy: 0.697 -> 0.739 | Improved |
| add level 3 on top of level 2 (ablation) | PARN (VGG-16) · Taniai benchmark (average) | matching accuracy: 0.739 -> 0.752 | Improved |
| add pixel-level module on top of level 3 (ablation) | PARN (VGG-16) · Taniai benchmark (average) | matching accuracy: 0.752 -> 0.760 | Improved |
| replace VGG-16 with ResNet-101 backbone | PARN · Taniai benchmark (average) | matching accuracy: 0.760 -> 0.788 | Improved |
| train level-1 module with semantic consistency supervision instead of synthetic deformations | PARN-Lv1 · Taniai benchmark (average) | matching accuracy: 0.644 -> 0.697 | Improved |
| add bilinear upsampling after each grid-level regression (ablation) | PARN · Taniai qualitative | artifact smoothness | Improved |
| use PARN instead of SCNet on PF-PASCAL at alpha=0.1 | SCNet · PF-PASCAL | PCK: 0.482 -> 0.491 | Improved |
| use PARN instead of SCNet on PF-PASCAL at alpha=0.05 | SCNet · PF-PASCAL | PCK: 0.260 -> 0.268 | Improved |
| use PARN instead of DCTM on Caltech-101 | DCTM · Caltech-101 | LT-ACC: 0.84 -> 0.87 | Improved |
| use PARN instead of DCTM on Caltech-101 | DCTM · Caltech-101 | LOC-ERR: 0.18 -> 0.21 | Hurt |
| use PARN instead of DCTM on Caltech-101 | DCTM · Caltech-101 | IoU: 0.53 -> 0.65 | Improved |
| use PARN instead of OHG on Taniai PASCAL subset | OHG · Taniai benchmark (PASCAL) | matching accuracy: 0.729 -> 0.688 | Hurt |
Later papers that compared with this method
1 later accepted paper reports PARN as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Show, Match and Segment: Joint Weakly Supervised Learning of Semantic Matching and Object Co-Segmentation (TPAMI 2020 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Pyramidal Semantic Correspondence Networks (TPAMI 2021 · similarity 0.93)
- Discrete-Continuous Transformation Matching for Dense Semantic Correspondence (TPAMI 2018 · similarity 0.88)
- Learning Semantic Correspondence Exploiting an Object-Level Prior (TPAMI 2020 · similarity 0.88)
- Recurrent Transformer Networks for Semantic Correspondence (NeurIPS 2018 · similarity 0.88)
- SFNet: Learning Object-Aware Semantic Correspondence (CVPR 2019 · similarity 0.88)
- Universal Correspondence Network (NeurIPS 2016 · similarity 0.87)
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.
