English | 简体中文
- Text detection
DBDetector
- [Optional] Direction classification
Classifer
is used to adjust the direction of images before text recognition - Character recognition
Recognizer
is used to recognize characters from images
According to different scenarios, FastDeploy provides the following OCR task deployment. Users need to download three models and dictionary files (or two, optional classifier) simultaneously to complete the entire OCR prediction process
The model download links in the following table are provided by PaddleOCR model library. Refer to PP-OCR Model List for details
OCR version | Text box detection | Direction classification model | Character recognition | Dictionary file | Note |
---|---|---|---|---|---|
ch_PP-OCRv3[Recommended] | ch_PP-OCRv3_det | ch_ppocr_mobile_v2.0_cls | ch_PP-OCRv3_rec | ppocr_keys_v1.txt | OCRv3 Original Ultra-Lightweight Model supports text detection in Chinese, English and multiple languages |
en_PP-OCRv3[Recommended] | en_PP-OCRv3_det | ch_ppocr_mobile_v2.0_cls | en_PP-OCRv3_rec | en_dict.txt | OCRv3 Original Ultra-Lightweight Model supports English and digital recognition. Its training data of detection model and recognition model is different from that of Chinese model, and no other differences can be detected |
ch_PP-OCRv2 | ch_PP-OCRv2_det | ch_ppocr_mobile_v2.0_cls | ch_PP-OCRv2_rec | ppocr_keys_v1.txt | OCRv2 Original Ultra-Lightweight Model supports text detection in Chinese, English and multiple languages |
ch_PP-OCRv2_mobile | ch_ppocr_mobile_v2.0_det | ch_ppocr_mobile_v2.0_cls | ch_ppocr_mobile_v2.0_rec | ppocr_keys_v1.txt | OCRv2 Original Ultra-Lightweight Model Supports text detection in Chinese, English and multiple languages with lighter weight than PPOCRv2 |
ch_PP-OCRv2_server | ch_ppocr_server_v2.0_det | ch_ppocr_mobile_v2.0_cls | ch_ppocr_server_v2.0_rec | ppocr_keys_v1.txt | OCRv2 Server Model supports text detection in Chinese, English and multiple languages. It has better effects though being larger than the ultra-lightweight model |