Research map / Image generation / GANs
Separating Content from Style Using Adversarial Learning for Recognizing Text in the Wild
Canjie Luo, Qingxiang Lin, Yuliang Liu, Lianwen Jin, Chunhua Shen
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 |
|---|---|---|---|
| generate clean images (separate content from background) as recognizer input | attention recognizer · CUTE | word accuracy (%): 74.3 -> 80.9 | Improved |
| generate clean images as recognizer input | attention recognizer · regular text (IIIT5K, SVT, IC03, IC13) | word accuracy (%): 92.2 -> 92.5 on IIIT5K, 94.0 -> 95.0 on IC03 | Improved |
| add style adversarial loss L_s (ablation) | content-supervised GAN · SVT-P | word accuracy (%): 71.9 -> 75.0 | Improved |
| add style adversarial loss L_s (ablation) | content-supervised GAN · IC15 | word accuracy (%): 65.5 -> 72.5 | Improved |
| add content classifier (content supervision) (ablation) | style-only GAN · SVT-P, CUTE, IC15 | word accuracy (%): Failed -> 75.0, 78.4, 72.5 | Improved |
| add feedback mechanism (supervise content classifier with recognizer predictions) (ablation) | GAN with L_s and L_c · SVT-P | word accuracy (%): 75.0 -> 79.2 | Improved |
| add feedback mechanism (ablation) | GAN with L_s and L_c · IC15 | word accuracy (%): 72.5 -> 73.0 | Improved |
| replace OTSU binarization with learned GAN separation | text normalization · SVT-P | word accuracy (%): 46.5 -> 79.2 | Improved |
| replace word-level CycleGAN with character-level adversarial generation | text normalization · SVT-P | word accuracy (%): 14.6 -> 79.2 | Improved |
| integrate method into ASTER | ASTER · CUTE | word accuracy (%): 79.9 -> 85.1 | Improved |
| integrate method into ESIR | ESIR · CUTE | word accuracy (%): 83.7 -> 87.8 | Improved |
| ensemble source and generated image predictions by confidence (Com.) | ASTER + Ours · CUTE | word accuracy (%): 85.1 -> 89.6 | Improved |
Closest papers by meaning
- DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022 · similarity 0.86)
- Adversarial Learning of Semantic Relevance in Text to Image Synthesis (AAAI 2019 · similarity 0.86)
- Dual Adversarial Inference for Text-to-Image Synthesis (ICCV 2019 · similarity 0.85)
- StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesis (ICML 2023 · similarity 0.85)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.85)
- Adversarial Text to Continuous Image Generation (CVPR 2024 · similarity 0.84)
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.
