Research map / Vision-language and multimodal / Human and embodied modeling
What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis
Jeonghun Baek, Geewook Kim, Junyeop Lee, Sungrae Park, Dongyoon Han, Sangdoo Yun, Seong Joon Oh, Hwalsuk Lee
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 TPS transformation (ablation) | STR framework (mean over combinations) · Regular benchmarks | accuracy (%): 85.6 -> 86.7 | Improved |
| add TPS transformation (ablation) | STR framework (mean over combinations) · Irregular benchmarks | accuracy (%): 65.7 -> 69.1 | Improved |
| replace VGG with ResNet feature extractor (ablation) | STR framework (mean over combinations) · Regular benchmarks | accuracy (%): 84.5 -> 88.3 | Improved |
| replace VGG with ResNet feature extractor (ablation) | STR framework (mean over combinations) · Irregular benchmarks | accuracy (%): 63.9 -> 71.0 | Improved |
| replace VGG with RCNN feature extractor (ablation) | STR framework (mean over combinations) · Regular benchmarks | accuracy (%): 84.5 -> 86.2 | Improved |
| replace VGG with RCNN feature extractor (ablation) | STR framework (mean over combinations) · Irregular benchmarks | accuracy (%): 63.9 -> 67.3 | Improved |
| add BiLSTM sequence modeling (ablation) | STR framework (mean over combinations) · Regular benchmarks | accuracy (%): 85.1 -> 87.6 | Improved |
| add BiLSTM sequence modeling (ablation) | STR framework (mean over combinations) · Irregular benchmarks | accuracy (%): 65.2 -> 69.7 | Improved |
| replace CTC with attention prediction (ablation) | STR framework (mean over combinations) · Regular benchmarks | accuracy (%): 85.5 -> 87.2 | Improved |
| replace CTC with attention prediction (ablation) | STR framework (mean over combinations) · Irregular benchmarks | accuracy (%): 66.1 -> 68.7 | Improved |
| train on MJSynth plus SynthText instead of MJSynth only (ablation) | Our best model · Unified benchmark (8,539 images) | total accuracy (%): 80.0 -> 84.1 | Improved |
| train on MJSynth plus SynthText instead of SynthText only (ablation) | Our best model · Unified benchmark (8,539 images) | total accuracy (%): 75.6 -> 84.1 | Improved |
Closest papers by meaning
- Revisiting Scene Text Recognition: A Data Perspective (ICCV 2023 · similarity 0.95)
- What if We Only Use Real Datasets for Scene Text Recognition? Toward Scene Text Recognition With Fewer Labels (CVPR 2021 · similarity 0.93)
- An Empirical Study of Scaling Law for Scene Text Recognition (CVPR 2024 · similarity 0.89)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.89)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.87)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.87)
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.
