Research map / Detection and segmentation / CNN features
Self-Attention Based Text Knowledge Mining for Text Detection
Qi Wan, Haoqin Ji, Linlin Shen
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 the Self-attention Decoder to STKM pre-training (ablation) | EAST · ICDAR2015 | F-measure: 80.02 -> 86.76; +6.74 | Improved |
| Add the Self-attention Decoder to STKM pre-training (ablation) | EAST · ICDAR2017 | F-measure: 62.36 -> 69.35; +6.99 | Improved |
| Add the Self-attention Decoder to STKM pre-training (ablation) | EAST · MSRA-TD500 | F-measure: 69.99 -> 79.33; +9.34 | Improved |
| Add the Self-attention Decoder to STKM pre-training (ablation) | PSENet · ICDAR2015 | F-measure: 80.82 -> 85.88; +5.06 | Improved |
| Add the Self-attention Decoder to STKM pre-training (ablation) | PSENet · CTW-1500 | F-measure: 76.89 -> 79.91; +3.02 | Improved |
| Add the Self-attention Decoder to STKM pre-training (ablation) | PSENet · TotalText | F-measure: 76.54 -> 82.15; +5.61 | Improved |
| Replace ImageNet pre-training with STKM pre-training | EAST · ICDAR2015 | F-measure: 80.87 -> 86.76; +5.89 | Improved |
| Replace ImageNet pre-training with STKM pre-training | EAST · ICDAR2017 | F-measure: 66.17 -> 69.35; +3.18 | Improved |
| Replace ImageNet pre-training with STKM pre-training | PSENet · ICDAR2015 | F-measure: 80.24 -> 85.88; +5.64 | Improved |
| Replace ImageNet pre-training with STKM pre-training | PSENet · CTW-1500 | F-measure: 78.74 -> 81.51; +2.77 | Improved |
| Use a PSENet-pretrained backbone instead of ImageNet pre-training | EAST · ICDAR2017 | F-measure: 66.17 -> 61.82; -4.35 | Hurt |
| Pre-train the backbone with SynthText | EAST · ICDAR2015 | F-measure: 80.87 -> 83.55; +2.68 | Improved |
Closest papers by meaning
- Single Shot Text Detector With Regional Attention (ICCV 2017 · similarity 0.87)
- Deep Short Text Classification with Knowledge Powered Attention (AAAI 2019 · similarity 0.86)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.86)
- Joint Visual Semantic Reasoning: Multi-Stage Decoder for Text Recognition (ICCV 2021 · similarity 0.85)
- Self-Supervised Learning for Text Recognition: A Critical Survey (IJCV 2025 · similarity 0.85)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.83)
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.
