Research map / Image generation / Text-to-image diffusion
IPAD: Iterative, Parallel, and Diffusion-Based Network for Scene Text Recognition
Xiaomeng Yang, Zhi Qiang Qiao, Yu Zhou
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 easy-first iterative decoding (ablation) | parallel decoder baseline · English synthetic | 7248-sample avg accuracy: 92.00 -> 92.47 | Improved |
| add easy-first iterative decoding (ablation) | parallel decoder baseline · BCTR Chinese | accuracy: 80.56 -> 82.78 | Improved |
| add discrete diffusion training (ablation) | IPAD w/o · English synthetic | 7248-sample avg accuracy: 92.47 -> 93.25 | Improved |
| add discrete diffusion training (ablation) | IPAD w/o · English real | 7248-sample avg accuracy: 95.85 -> 96.39 | Improved |
| add discrete diffusion training (ablation) | IPAD w/o · BCTR Chinese | accuracy: 82.78 -> 84.08 | Improved |
| add mimicking learning instead of diffusion (ablation) | IPAD w/o · English real | 7248-sample avg accuracy: 95.85 -> 95.94 | No clear effect |
| replace CNN encoder with ViT encoder (ablation) | PIMNet · Six English benchmarks | avg accuracy: 90.5 -> 92.6 | Improved |
| add mimicking learning with ViT encoder (ablation) | IPAD w/o · Six English benchmarks | avg accuracy: 92.5 -> 92.6 | No clear effect |
| add timestep embedding (sinusoidal T=25) (ablation) | IPAD · BCTR Chinese | accuracy: 81.97 -> 84.08 | Improved |
| use sinusoidal timestep embedding instead of learnable embedding (ablation) | IPAD · BCTR Chinese | accuracy: 82.65 -> 84.08 | Improved |
| increase easy-first iterations from 1 to 5 (ablation) | IPAD · English synthetic | avg accuracy: 92.15 -> 93.25 | Improved |
| increase iterations from 5 to 25 (ablation) | IPAD · English real | avg accuracy: 96.39 -> 96.27 | No clear effect |
Closest papers by meaning
- Context Perception Parallel Decoder for Scene Text Recognition (TPAMI 2025 · similarity 0.88)
- ABINet++: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Spotting (TPAMI 2022 · similarity 0.88)
- Scene Text Recognition from Two-Dimensional Perspective (AAAI 2019 · similarity 0.88)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- Brush Your Text: Synthesize Any Scene Text on Images via Diffusion Model (AAAI 2024 · similarity 0.86)
- Image as a Language: Revisiting Scene Text Recognition via Balanced, Unified and Synchronized Vision-Language Reasoning Network (AAAI 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.
