Research map / Detection and segmentation / CNN features
Learning Shape-Aware Embedding for Scene Text Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Shape-Aware Loss instead of Disc Loss (ablation) | scene text detection model · MSRA-TD500 | H-mean: 81.1 -> 82.9 | Helped |
| Add Shape-Aware Loss instead of Disc Loss (ablation) | scene text detection model · ICDAR15 | H-mean: 78.0 -> 82.2 | Helped |
| Use embedding clustering with the full and center segmentation maps instead of only dilating Center Map boxes (ablation) | proposed detection pipeline · ICDAR15 (1760) | H-mean: 84.3 -> 86.6 | Helped |
| Use embedding clustering with the full and center segmentation maps instead of only enlarging Center Map boxes (ablation) | proposed detection pipeline · MSRA-TD500 | H-mean: 77.0 -> 82.9 | Helped |
| Keep separate embedding and segmentation branches instead of merging them into one branch (ablation) | proposed network · ICDAR15 | H-mean: 79.2 -> 84.8 | Helped |
| Keep the Full Map rather than replacing it with dilated Center Map (ablation) | proposed network · ICDAR15 | H-mean: 69.1 -> 84.8 | Helped |
| Use embedding clustering to generate text boxes rather than direct Center Map box dilation (ablation) | proposed detection pipeline · ICDAR15 (1280) | H-mean: 81.3 -> 84.8 | Helped |
| Increase the ICDAR15 test input scale from 1280 to 1760 (ablation) | proposed method · ICDAR15 | H-mean: 84.8 -> 86.6 | Helped |
| Use Shape-Aware Loss rather than Disc Loss (ablation) | embedding branch · MSRA-TD500 | inter embedding distance: 1.7 -> 1.9 | Helped |
| Use Shape-Aware Loss rather than Disc Loss (ablation) | embedding branch · MSRA-TD500 | intra embedding distance: 0.5 -> 0.4 | Helped |
| Directly cluster embedding features rather than use the proposed segmentation-guided cluster pipeline (ablation) | proposed detection pipeline · MSRA-TD500 | H-mean | Hurt |
Closest papers by meaning
- All You Need Is Boundary: Toward Arbitrary-Shaped Text Spotting (AAAI 2020 · similarity 0.89)
- PixelLink: Detecting Scene Text via Instance Segmentation (AAAI 2018 · similarity 0.89)
- Geometry-Aware Scene Text Detection With Instance Transformation Network (CVPR 2018 · similarity 0.89)
- LRANet: Towards Accurate and Efficient Scene Text Detection with Low-Rank Approximation Network (AAAI 2024 · similarity 0.88)
- Kernel Adaptive Convolution for Scene Text Detection via Distance Map Prediction (CVPR 2024 · similarity 0.87)
- Towards Unified Multi-granularity Text Detection with Interactive Attention (ICML 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.