PP-OCR series model comparison
PaddleOCR/ppocr_introduction.md at release/2.6 · PaddlePaddle/PaddleOCR · GitHub
(1) PP-OCR: Two-stage OCR system, in which DB is selected as the text detection algorithm, CRNN is selected as the text recognition algorithm , and a text orientation classifier is added between the detection and recognition modules .
(2) PP-OCRv2: Based on PP-OCR key optimization in five aspects, the detection model adopts CML collaborative mutual learning knowledge distillation strategy and CopyPaste data augmentation strategy; the recognition model adopts LCNet lightweight backbone network and UDML improved knowledge distillation strategy And the Enhanced CTC loss loss function improvement (as shown in the red box above), has further achieved significant improvements in inference speed and prediction effect.
(3) PP-OCRv3: Upgraded in 9 aspects based on PP-OCRv2.
Model download addresses and configuration files for each version
PaddleOCR/models_list.md at release/2.6 · PaddlePaddle/PaddleOCR · GitHub
ch_ppocr_server_v2.0_det：
Training model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar
Inference model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar
ch_ppocr_mobile_v2.0_det：
Training model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar
Inference model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
ch_ppocr_server_v2.0_rec：
Training model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar
Inference model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar
ch_ppocr_mobile_v2.0_rec：
Training model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar
Inference model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
ch_ppocr_mobile_v2.0_cls：
https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar pre-training model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar
Inference model: https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
PP-OCR inference model uses
详见：PaddleOCR/quickstart.md at static · PaddlePaddle/PaddleOCR · GitHub
git clone https://github.com/PaddlePaddle/PaddleOCR.gitCreate a new inference folder, put the downloaded inference model in and unzip it:
(1) Test the effect of the det reasoning model:
python tools/infer/predict_det.py --image_dir doc/imgs/00111002.jpg --det_model_dir inference/ch_ppocr_server_v2.0_det_inferResults are stored under inference_results: det_res_00111002.jpg and det_results.txt
(2) Test the effect of the cls reasoning model:
python tools/infer/predict_cls.py --image_dir doc/imgs_words/ch/word_4.jpg --cls_model_dir inference/ch_ppocr_mobile_v2.0_cls_inferResult: Predicts of doc/imgs_words/ch/word_4.jpg:['0', 0.9999982]
The first is the angle, the second is the confidence
(3) Test the effect of the rec reasoning model:
python tools/infer/predict_rec.py --image_dir doc/imgs_words/ch/word_4.jpg --rec_model_dir inference/ch_ppocr_server_v2.0_rec_inferppocr INFO: Predicts of doc/imgs_words/ch/word_4.jpg:('handed in from live tool', 0.2635100483894348)
The first is the literal result, the second is the confidence
(4) combined
python tools/infer/predict_system.py --image_dir doc/imgs/00111002.jpg \
--det_model_dir inference/ch_ppocr_server_v2.0_det_infer/ \
--rec_model_dir inference/ch_ppocr_server_v2.0_rec_infer/ \
--cls_model_dir inference/ch_ppocr_mobile_v2.0_cls_infer/ \
--use_angle_cls True \
--use_space_char TrueResults are stored under inference_results: 00111002.jpg and system_results.txt
PP-OCRv3 improvement points:
PaddleOCR/PP-OCRv3_introduction.md at release/2.6 · PaddlePaddle/PaddleOCR · GitHub
PP-OCRv3 detection model training steps reference:
PaddleOCR/PP-OCRv3_det_train.md at release/2.6 · PaddlePaddle/PaddleOCR · GitHub