Research map / Detection and segmentation / CNN features
Learning to Compose Hypercolumns for Visual Correspondence
Juhong Min, Jongmin Lee, Jean Ponce, Minsu Cho
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 conv feature transformation with channel reduction (ablation) | DHPF · PF-PASCAL | PCK@0.1: 74.7 -> 90.7 | Improved |
| add probabilistic Hough matching (ablation) | DHPF · PF-PASCAL | PCK@0.1: 86.9 -> 90.7 | Improved |
| add Gumbel layer gating (ablation) | DHPF · PF-PASCAL | PCK@0.1: 88.1 -> 90.7 | Improved |
| add keypoint importance weight omega_m (ablation) | DHPF · PF-PASCAL | PCK@0.1: 86.1 -> 90.7 | Improved |
| add layer selection loss (ablation) | DHPF · PF-PASCAL | PCK@0.1: 89.2 -> 90.7 | Improved |
| lower layer selection rate mu from 0.5 to 0.3 (ablation) | DHPF · PF-PASCAL | PCK@0.1 / time: 90.7 -> 88.7 | Mixed |
| lower layer selection rate mu from 0.5 to 0.4 (ablation) | DHPF · PF-PASCAL | PCK@0.1 / time: 90.7 -> 90.4 | Mixed |
| use Gumbel hard gating instead of sigmoid soft gating (ablation) | DHPF · PF-PASCAL | PCK@0.1: 88.2 -> 90.7 | Improved |
| use Gumbel gating instead of sigmoid + l1 sparsity (ablation) | DHPF · PF-PASCAL | PCK@0.1: 87.2 -> 90.7 | Improved |
| use weak supervision for training on SPair-71k (fine-tune instead of transfer) | DHPF · SPair-71k | PCK: 28.5 -> 27.7 | Hurt |
| strong supervision with ResNet-101 vs previous state of the art HPF | DHPF vs HPF · PF-PASCAL | PCK@0.1: 84.8 -> 90.7 | Improved |
| use ResNet-101 instead of ResNet-50 backbone | DHPF · PF-PASCAL (strong) | PCK@0.1: 88.9 -> 90.7 | Improved |
Closest papers by meaning
- Hyperpixel Flow: Semantic Correspondence With Multi-Layer Neural Features (ICCV 2019 · similarity 0.92)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.89)
- Dynamic Context Correspondence Network for Semantic Alignment (ICCV 2019 · similarity 0.88)
- Universal Correspondence Network (NeurIPS 2016 · similarity 0.88)
- Pixel-level Semantic Correspondence through Layout-aware Representation Learning and Multi-scale Matching Integration (CVPR 2024 · similarity 0.86)
- Unifying Feature and Cost Aggregation with Transformers for Semantic and Visual Correspondence (ICLR 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.
