Research map / Vision-language and multimodal / Human and embodied modeling
TextOCR: Towards Large-Scale End-to-End Reasoning for Arbitrary-Shaped Scene Text
Amanpreet Singh, Guan Pang, Mandy Toh, Jing Huang, Wojciech Galuba, Tal Hassner
Check 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 |
|---|---|---|---|
| Train the recognizer on TextOCR instead of COCOText | TPS-ResNet-BiLSTM-Attn · TextOCR | word accuracy: 43.20 -> 65.65 | Improved |
| Fine-tune STAR-Net with TextOCR | STAR-Net · ICDAR2015 | word accuracy: 75.81 -> 86.09 | Improved |
| Fine-tune TPS-ResNet-BiLSTM-Attn with TextOCR | TPS-ResNet-BiLSTM-Attn · COCOText | word accuracy: 56.34 -> 69.15 | Improved |
| Fine-tune Mask TextSpotter v3 with TextOCR | Mask TextSpotter v3 · Total-Text (None) | F-measure: 71.2 -> 74.5 | Improved |
| Use TextOCR-trained OCR and MTS v3 LH features in place of M4C's Rosetta/OCR setup | PixelM4C · TextVQA val | accuracy (%): 39.40 -> 41.23 | Improved |
| Add ST-VQA training data (ablation) | PixelM4C with TextOCR OCR and MTS v3 LH features · TextVQA val | accuracy (%): 41.23 -> 42.12 | Improved |
| Replace MTS v3 LH features with MTS v3 fc7 features (ablation) | PixelM4C with TextOCR OCR · TextVQA val | accuracy (%): 41.23 -> 39.64 | Hurt |
| Use TextOCR annotations directly instead of recognized OCR tokens (ablation) | PixelM4C · TextVQA val | accuracy (%): 41.23 -> 46.36 | Improved |
| Increase OCR token limit from 50 to 100 (ablation) | PixelM4C with TextOCR annotations · TextVQA val | accuracy (%): 45.22 -> 46.42 | Improved |
| Increase OCR token limit from 100 to 200 (ablation) | PixelM4C with TextOCR-trained OCR · TextVQA val | accuracy (%): 39.41 -> 39.41 | No clear effect |
| Use TextOCR-trained OCR tokens for TextCaps instead of Rosetta tokens | PixelM4C-Captioner · TextCaps val | CIDEr: 89.60 -> 85.32 | Hurt |
Closest papers by meaning
- UPOCR: Towards Unified Pixel-Level OCR Interface (ICML 2024 · similarity 0.89)
- MME-VideoOCR: Evaluating OCR-Based Capabilities of Multimodal LLMs in Video Scenarios (NeurIPS 2025 · similarity 0.87)
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning (NeurIPS 2025 · similarity 0.87)
- OCR-Reasoning Benchmark: Unveiling the True Capabilities of MLLMs in Complex Text-Rich Image Reasoning (ICLR 2026 · similarity 0.87)
- ODM: A Text-Image Further Alignment Pre-training Approach for Scene Text Detection and Spotting (CVPR 2024 · similarity 0.87)
- InstructOCR: Instruction Boosting Scene Text Spotting (AAAI 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.
