Adaptive Multi-Scale Feature Fusion for Paragraph-Level Handwritten Text Recognition via Spatial Attention

Computer Science

• Annotation 50%
• Attention (Machine Learning) 50%
• Convolutional Neural Network 100%
• Feature Extraction 50%
• Feature Fusion 100%
• Handwriting Recognition 100%
• local feature 50%
• Semantic Dependency 50%
• Spatial Location 50%
• Vision Transformer 50%

Keyphrases

• Adaptive multi-scale Fusion 33%
• Autoregressive Transformer 33%
• External Lexicon 33%
• Feature Consistency Loss 33%
• Global Context Modeling 33%
• Lightweight CNN 33%
• Multiple Lines 33%
• Recurrence-free 33%
• Strong Sequences 33%