Early fusion of image patches and pen-stroke tokens in a transformer reaches 96.4% accuracy on isolated IAMOn-DB characters, slightly above the prior dual-input system.
SN Computer Science2(1), 20 (2021)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A Transformer Based Handwriting Recognition System Jointly Using Online and Offline Features
Early fusion of image patches and pen-stroke tokens in a transformer reaches 96.4% accuracy on isolated IAMOn-DB characters, slightly above the prior dual-input system.