Research map / Detection and segmentation / CNN features
Chinese Street View Text: Large-Scale Chinese Text Reading With Partially Supervised Learning
Yipeng Sun, Jiaming Liu, Wei Liu, Junyu Han, Errui Ding, Jingtuo Liu
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 400K weakly annotated images with online proposal matching (partially supervised learning) (ablation) | End2End · C-SVT test | end-to-end F-score: 28.58 -> 33.88 | Improved |
| add 400K weakly annotated images with online proposal matching (ablation) | End2End · C-SVT test | end-to-end AED: 19.68 -> 17.59 | Improved |
| add 400K weak annotations instead of equal-cost 4.4K extra full annotations (ablation) | End2End · C-SVT test | end-to-end F-score: 29.85 -> 33.88 | Improved |
| increase weak annotations from 25K to 400K (ablation) | End2End-PSL · C-SVT test | end-to-end F-score: 31.46 -> 33.88 | Improved |
| add 400K weak annotations to ICDAR 2017-RCTW training (ablation) | End2End · ICDAR 2017-RCTW | end-to-end AED: 27.5 -> 24.2 | Improved |
| add 400K weak annotations with multi-scale testing (ablation) | End2End-MS · ICDAR 2017-RCTW | end-to-end AED: 26.2 -> 22.1 | Improved |
| add 400K weak annotations (ablation) | End2End · ICDAR 2017-RCTW | detection F-score: 60.1 -> 60.2 | No clear effect |
| add 400K weak annotations (ablation) | End2end recognition branch · C-SVT test (GT boxes) | recognition accuracy: 40.20 -> 48.33 | Improved |
| raise OPM threshold tau from 0.05 to 0.2 (ablation) | OPM · 500 weak images, IoU 0.5 | recall: 45 -> 66 | Improved |
| raise OPM threshold tau from 0.05 to 0.2 (ablation) | OPM · 500 weak images, IoU 0.5 | precision: 98 -> 92 | Hurt |
| add 400K weak annotations on English data (COCO-Text) (ablation) | End2End · ICDAR 2015 end-to-end weak lexicon | F-score: 75.09 -> 76.25 | Improved |
| train with C-SVT full annotations instead of CTW | End2End · ICDAR 2017-RCTW | end-to-end AED: 33.6 -> 26.8 | Improved |
Closest papers by meaning
- Beyond Cropped Regions: New Benchmark and Corresponding Baseline for Chinese Scene Text Retrieval in Diverse Layouts (ICML 2025 · similarity 0.87)
- PageNet: Towards End-to-End Weakly Supervised Page-Level Handwritten Chinese Text Recognition (IJCV 2022 · similarity 0.86)
- CNVid-3.5M: Build, Filter, and Pre-Train the Large-Scale Public Chinese Video-Text Dataset (CVPR 2023 · similarity 0.85)
- BTS: A Bi-Lingual Benchmark for Text Segmentation in the Wild (CVPR 2022 · similarity 0.85)
- A Benchmark for Chinese-English Scene Text Image Super-Resolution (ICCV 2023 · similarity 0.84)
- CVLUE: A New Benchmark Dataset for Chinese Vision-Language Understanding Evaluation (AAAI 2025 · 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.
