AI Paper Atlas

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

CVPR 2021 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded scene text and structure · 157 citations (OpenAlex)

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 changedWhereMeasuredEffect
Train the recognizer on TextOCR instead of COCOTextTPS-ResNet-BiLSTM-Attn · TextOCRword accuracy: 43.20 -> 65.65Improved
Fine-tune STAR-Net with TextOCRSTAR-Net · ICDAR2015word accuracy: 75.81 -> 86.09Improved
Fine-tune TPS-ResNet-BiLSTM-Attn with TextOCRTPS-ResNet-BiLSTM-Attn · COCOTextword accuracy: 56.34 -> 69.15Improved
Fine-tune Mask TextSpotter v3 with TextOCRMask TextSpotter v3 · Total-Text (None)F-measure: 71.2 -> 74.5Improved
Use TextOCR-trained OCR and MTS v3 LH features in place of M4C's Rosetta/OCR setupPixelM4C · TextVQA valaccuracy (%): 39.40 -> 41.23Improved
Add ST-VQA training data (ablation)PixelM4C with TextOCR OCR and MTS v3 LH features · TextVQA valaccuracy (%): 41.23 -> 42.12Improved
Replace MTS v3 LH features with MTS v3 fc7 features (ablation)PixelM4C with TextOCR OCR · TextVQA valaccuracy (%): 41.23 -> 39.64Hurt
Use TextOCR annotations directly instead of recognized OCR tokens (ablation)PixelM4C · TextVQA valaccuracy (%): 41.23 -> 46.36Improved
Increase OCR token limit from 50 to 100 (ablation)PixelM4C with TextOCR annotations · TextVQA valaccuracy (%): 45.22 -> 46.42Improved
Increase OCR token limit from 100 to 200 (ablation)PixelM4C with TextOCR-trained OCR · TextVQA valaccuracy (%): 39.41 -> 39.41No clear effect
Use TextOCR-trained OCR tokens for TextCaps instead of Rosetta tokensPixelM4C-Captioner · TextCaps valCIDEr: 89.60 -> 85.32Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy