History

yhq 397ecf2658 Feature/paddleocr inference (#148 ) * add ocr model and convert weights from paddleocrv3		2022-09-28 14:03:16 +08:00
..
detection	Feature/paddleocr inference (#148 )	2022-09-28 14:03:16 +08:00
direction	Feature/paddleocr inference (#148 )	2022-09-28 14:03:16 +08:00
recognition	Feature/paddleocr inference (#148 )	2022-09-28 14:03:16 +08:00
README.md	Feature/paddleocr inference (#148 )	2022-09-28 14:03:16 +08:00

README.md

OCR algorithm

PP-OCRv3

We convert PaddleOCRv3 models to pytorch style, and provide end2end interface to recognize text in images, by simplely load exported models.

detection

We test on on icdar2015 dataset.

Algorithm	backbone	configs	precison	recall	Hmean	Download
DB	MobileNetv3	det_model_en.py	0.7803	0.7250	0.7516	log-model
DB	R50	det_model_en_r50.py	0.8622	0.8218	0.8415	log-model

recognition

We test on on DTRB dataset.

Algorithm	backbone	configs	acc	Download
SVTR	MobileNetv1	rec_model_en.py	0.7536	log-model

predict

We provide exported models contain weights and process config for easyly predict, which convert from PaddleOCRv3.

detection model

language	Download
chinese	chinese_det.pth
english	english_det.pth
multilingual	multilingual_det.pth

recognition model

language	Download
chiese	chinese_rec.pth
english	english_rec.pth
korean	korean_rec.pth
japan	japan_rec.pth
chinese_cht	chinese_cht_rec.pth
Telugu	te_rec.pth
Canada	ka_rec.pth
Tamil	ta_rec.pth
latin	latin_rec.pth
cyrillic	cyrillic_rec.pth
devanagari	devanagari_rec.pth

direction model

language	Download
	direction.pth

usage

detection

import cv2
from easycv.predictors.ocr import OCRDetPredictor
predictor = OCRDetPredictor(model_path)
out = predictor([img_path]) # out = predictor([img])
img = cv2.imread(img_path)
out_img = predictor.show_result(out[0], img)
cv2.imwrite(out_img_path,out_img)

recognition

import cv2
from easycv.predictors.ocr import OCRRecPredictor
predictor = OCRRecPredictor(model_path)
out = predictor([img_path])
print(out)

end2end

import cv2
from easycv.predictors.ocr import OCRPredictor
! wget http://pai-vision-data-hz.oss-cn-zhangjiakou.aliyuncs.com/EasyCV/modelzoo/ocr/simfang.ttf
! wget http://pai-vision-data-hz.oss-cn-zhangjiakou.aliyuncs.com/EasyCV/modelzoo/ocr/ocr_det.jpg
predictor = OCRPredictor(
    det_model_path=path_to_detmodel,
    rec_model_path=path_to_recmodel,
    cls_model_path=path_to_clsmodel,
    use_angle_cls=True)
filter_boxes, filter_rec_res = predictor(img_path)
img = cv2.imread('ocr_det.jpg')
out_img = predictor.show(
    filter_boxes[0],
    filter_rec_res[0],
    img,
    font_path='simfang.ttf')
cv2.imwrite('out_img.jpg', out_img)

There are some ocr results.