chinese-ocr-webapp

OCR webapp that uses YOLOv5 to detect chinese characters, ConvNet to classify detected characters, and Transformer network to translate Chinese to English.

Example

Input Image:

Output of YOLOv5:

Output of CNN:
['你', '好']
Output of Transformer:
Hi

Character Detection with YOLOv5

YOLOv5 achieved a 70.8 mAP after 100 epochs.

Character Classification with CNN

best model: 90% accuracy on dev set after 20 epochs

Chinese to English Translation with Transformer

best model - epoch: 40 val loss: 0.4836350381374359 bleu: 0.7384979341626325