AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Perceiving Stroke-Semantic Context: Hierarchical Contrastive Learning for Robust Scene Text Recognition

Hao Liu, Bin Wang, Zhimin Bao, Mobai Xue, Sheng Kang, Deqiang Jiang, Yinsong Liu, Bo Ren

AAAI 2022 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Empirical contrastive visual learning · 46 citations (OpenAlex)

Read the paper

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
Use hierarchical stroke- and semantic-context contrastive learning instead of cross-sample sequential contrastScene text recognition representations · IIIT5K, IC03, IC13word accuracy: 35.7 -> 43.4; 43.6 -> 50.6; 43.5 -> 51.2Improved
Pre-train PerSec on UTI-100MPerSec-ViT · IIIT5K, IC03, IC13word accuracy: 38.4 -> 43.4; 46.2 -> 50.6; 46.7 -> 51.2Improved
Use a ViT-based rather than CNN-based encoderPerSec · IIIT5K, IC03, IC13word accuracy: 39.2 -> 43.4; 47.8 -> 50.6; 48.2 -> 51.2Improved
Add the stroke context perceiver (ablation)PerSec-CNN · IC13word accuracy: 86.3 -> 87.9; 86.0 -> 87.9Improved
Add the stroke context perceiver (ablation)PerSec-ViT · IC13word accuracy: 87.5 -> 89.7; 88.0 -> 89.2Improved
Add the semantic context perceiver (ablation)PerSec-CNN · IC13word accuracy: 86.0 -> 87.9; 86.7 -> 88.9Improved
Add the semantic context perceiver (ablation)PerSec-ViT · IC13word accuracy: 87.2 -> 89.7; 87.9 -> 89.2Improved
Remove both stroke and semantic context perceivers (ablation)PerSec-CNN · IC13word accuracy: 83.3 -> 87.9; 85.4 -> 88.9Hurt
Remove both stroke and semantic context perceivers (ablation)PerSec-ViT · IC13word accuracy: 84.2 -> 89.7; 86.2 -> 89.2Hurt
Remove the quantizer and contrast input features directly with context-aggregated features (ablation)PerSec · IC13convergenceHurt
Use PerSec-pretrained encoders for existing recognizersRobustScanner, SATRN, SAR, and NRTR · IC13, IIIT5K, SVT, IC15, SVTP, CT80word accuracy: 92.8 -> 95.1; 94.6 -> 95.2; 88.1 -> 91.2; 76.9 -> 78.1; 79.5 -> 81.2; 89.7 -> 91.3Improved

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