Web Server와 연동할 AI Server(OCR)
- text_recognition 모델 (accuracy 95%, 문자최대길이 20자 이내(띄어쓰기 포함) 학습모델) https://drive.google.com/file/d/1W7Y90G3YFED75D1wy-cI6t11QwhXaR-f/view?usp=share_link
- text_detection 모델(craft_mlt_25k/pretrained 모델) https://drive.google.com/file/d/1LG_UxK_dwMagXZHg4fcPia6LoiAjTJqM/view?usp=sharing
https://drive.google.com/drive/folders/1YuA5CrXsoS9WYELPSsJTOSfpxk0l_Qsb?usp=sharing
-
SwinIR GPP loss, ligtweight, grayscale(ours)
저해상도(세로 길이 16~32) 이미지 text recognition accuracy 86.19% (원본 이미지 성능 79.37%)
https://drive.google.com/file/d/1mZfp_FFZzBD_SitDFw05O0BtFo8s0sPq/view?usp=share_link -
SwinIR pretrined model(for RGB Image, pretrained model)
저해상도(세로 길이 16~32) 이미지 text recognition accuracy 83.8% (원본 이미지 성능 79.37%)
https://drive.google.com/file/d/1fljDr7p70ZVovVE7tk4rFua_blceen-0/view?usp=share_link
- CNN Document Image Classification (VGG16 backbone + classifier head) - acc: 0.97 loss: 0.14 https://drive.google.com/file/d/1oD6L3yco63L8uksK2pL8DXGrrRUb8pyt/view
python manage.py runserver