Research map / Detection and segmentation / CNN features
WordSup: Exploiting Word Annotations for Character Based Text Detection
Han Hu, Chengquan Zhang, Yuxuan Luo, Yuzhuo Wang, Junyu Han, Errui Ding
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 1k real ICDAR15 training images with word supervision to 50k synthetic images (ablation) | WordSup VGG16 character model · ICDAR15 | F-measure: 69.18% -> 78.16% | Improved |
| add 1k real ICDAR15 images with word supervision (ablation) | WordSup VGG16 character model · ICDAR13 (DetEval) | F-measure: 86.92 -> 90.34 | Improved |
| add COCO-Text real images with word supervision to synthetic training (ablation) | WordSup VGG16 character model · COCO-Text | F-measure: 32.5 -> 36.8 | Improved |
| train with 48,000 word-supervised images in addition to 2,000 character-annotated images instead of 2,000 characters alone (ablation) | character detector · VGG SynthText-part (character detection) | character detection precision-recall | Improved |
| use word supervision for 48k images instead of full character supervision of 50k (ablation) | character detector · VGG SynthText-part (character detection) | character detection precision-recall | No clear effect |
| add second-level hard negative mining (ablation) | basic character detector · VGG SynthText-part (character detection) | character detection precision-recall | Improved |
| use 1/4 resolution feature maps merged with deeper cues instead of 1/16 as in general detectors | character detector vs Faster RCNN and SSD · VGG SynthText-part (character detection) | character detection precision-recall | Improved |
Closest papers by meaning
- Character Region Awareness for Text Detection (CVPR 2019 · similarity 0.91)
- WeText: Scene Text Detection Under Weak Supervision (ICCV 2017 · similarity 0.89)
- Language Matters: A Weakly Supervised Vision-Language Pre-training Approach for Scene Text Detection and Spotting (ECCV 2022 · similarity 0.89)
- General Detection-based Text Line Recognition (NeurIPS 2024 · similarity 0.88)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.86)
- LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 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.
