OCR webapp that uses YOLOv5 to detect chinese characters, ConvNet to classify detected characters, and Transformer network to translate Chinese to English.
Input Image:
Output of YOLOv5:
Output of CNN:
['你', '好']
Output of Transformer:
Hi
YOLOv5 achieved a 70.8 mAP after 100 epochs.
best model: 90% accuracy on dev set after 20 epochs
best model - epoch: 40 val loss: 0.4836350381374359 bleu: 0.7384979341626325