Handwritten Text Recognition On Iam

Metrics

CER

WER

Results

Performance results of various models on this benchmark

			Paper Title
Transformer w/ CNN	7.62	-	Pay Attention to What You Read: Non-recurrent Handwritten Text-Line Recognition
FPHR Paragraph Level (~145 dpi)	6.7	-	Full Page Handwriting Recognition via Image to Sequence Extraction
Leaky LP Cell	6.6	15.9	No Padding Please: Efficient Neural Handwriting Recognition
FPHR+Aug Line Level (~145 dpi)	6.5	-	Full Page Handwriting Recognition via Image to Sequence Extraction
Decouple Attention Network	6.4	19.6	Decoupled Attention Network for Text Recognition
Start, Follow, Read	6.4	23.2	Start, Follow, Read: End-to-End Full-Page Handwriting Recognition
FPHR+Aug Paragraph Level (~145 dpi)	6.3	-	Full Page Handwriting Recognition via Image to Sequence Extraction
Easter2.0	6.21	-	Easter2.0: Improving convolutional models for handwritten text recognition
LSTM with attention	4.87	-	Evaluating Sequence-to-Sequence Models for Handwritten Text Recognition
HTR-VT(line-level)	4.7	14.9	HTR-VT: Handwritten Text Recognition with Vision Transformer
Transformer w/ CNN (+synth)	4.67	-	Pay Attention to What You Read: Non-recurrent Handwritten Text-Line Recognition
VAN	4.32	16.24	End-to-end Handwritten Paragraph Text Recognition Using a Vertical Attention Network
TrOCR-small 62M	4.22	-	TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models
TrOCR-base 334M	3.42	-	TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models
Transformer + CNN	2.96	-	Rethinking Text Line Recognition Models
TrOCR-large 558M	2.89	-	TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models
Self-Attention + CTC + language model	2.75	-	Rethinking Text Line Recognition Models
DTrOCR 105M	2.38	-	DTrOCR: Decoder-only Transformer for Optical Character Recognition

0 of 18 row(s) selected.

Handwritten Text Recognition On Iam | SOTA | HyperAI