8. Results

Results

Configurations for model training:

C1: Mixed Training

C2: Synthetic Only Training + Real Finetune

C3: Mixed Training + Real Finetune

Model	Training Config.	CER(%)	WER(%)
Attention LSTM	C3	3.86	13.86
Attention LSTM	C2	4.59	17.97
Attention LSTM	C1	4.77	16.84
CNN-RNN	C3	4.72	22.53
CNN-RNN	C2	4.78	22.97
CNN-RNN	C1	4.83	23.04
Ind.senz	-	18.15	35.72
Google OCR	-	12.82	40.63
Tesseract	-	17.27	66.18