产品信息|时会_PaddleOCR之安装测试

篇首语：本文由编程笔记#小编为大家整理，主要介绍了PaddleOCR之安装测试相关的知识，希望对你有一定的参考价值。

paddleocr 号称最强&＃xff0c;怀着好奇心&＃xff0c;对该软件进行一个安装测试

创建一个虚拟环境

conda create -n paddle_env python&＃61;3.8

进入环境&＃xff0c;安装 gpu 版本的 paddlepaddle

conda activate paddle_env
python3 -m pip install paddlepaddle-gpu -i https://mirror.baidu.com/pypi/simple

再安装 paddleocr

pip install "paddleocr>&＃61;2.0.1"

注意&＃xff1a;使用 whl 包安装的库支持的算法有限&＃xff0c;比如文本检测部分只支持 DB 算法&＃xff0c;如果要使用其它模型&＃xff0c;需要通过源码安装调试&＃xff0c;具体方法在后续文章中会介绍 

下载一张测试图片进行测试&＃xff0c;输出如下

$ paddleocr --image_dir ./word_1.jpg --use_angle_cls true
[2021/12/08 15:15:57] root WARNING: version PP-OCRv2 not support cls models, auto switch to version PP-OCR
Namespace(benchmark&＃61;False, cls_batch_num&＃61;6, cls_image_shape&＃61;&＃39;3, 48, 192&＃39;, cls_model_dir&＃61;&＃39;/home/lthpc/.paddleocr/2.3.0.2/ocr/cls/ch_ppocr_mobile_v2.0_cls_infer&＃39;, cls_thresh&＃61;0.9, cpu_threads&＃61;10, det&＃61;True, det_algorithm&＃61;&＃39;DB&＃39;, det_db_box_thresh&＃61;0.6, det_db_score_mode&＃61;&＃39;fast&＃39;, det_db_thresh&＃61;0.3, det_db_unclip_ratio&＃61;1.5, det_east_cover_thresh&＃61;0.1, det_east_nms_thresh&＃61;0.2, det_east_score_thresh&＃61;0.8, det_limit_side_len&＃61;960, det_limit_type&＃61;&＃39;max&＃39;, det_model_dir&＃61;&＃39;/home/lthpc/.paddleocr/2.3.0.2/ocr/det/ch/ch_PP-OCRv2_det_infer&＃39;, det_pse_box_thresh&＃61;0.85, det_pse_box_type&＃61;&＃39;box&＃39;, det_pse_min_area&＃61;16, det_pse_scale&＃61;1, det_pse_thresh&＃61;0, det_sast_nms_thresh&＃61;0.2, det_sast_polygon&＃61;False, det_sast_score_thresh&＃61;0.5, drop_score&＃61;0.5, e2e_algorithm&＃61;&＃39;PGNet&＃39;, e2e_char_dict_path&＃61;&＃39;./ppocr/utils/ic15_dict.txt&＃39;, e2e_limit_side_len&＃61;768, e2e_limit_type&＃61;&＃39;max&＃39;, e2e_model_dir&＃61;None, e2e_pgnet_mode&＃61;&＃39;fast&＃39;, e2e_pgnet_polygon&＃61;True, e2e_pgnet_score_thresh&＃61;0.5, e2e_pgnet_valid_set&＃61;&＃39;totaltext&＃39;, enable_mkldnn&＃61;False, gpu_mem&＃61;500, help&＃61;&＃39;&＃61;&＃61;SUPPRESS&＃61;&＃61;&＃39;, image_dir&＃61;&＃39;./word_1.jpg&＃39;, ir_optim&＃61;True, label_list&＃61;[&＃39;0&＃39;, &＃39;180&＃39;], lang&＃61;&＃39;ch&＃39;, layout_path_model&＃61;&＃39;lp://PubLayNet/ppyolov2_r50vd_dcn_365e_publaynet/config&＃39;, max_batch_size&＃61;10, max_text_length&＃61;25, min_subgraph_size&＃61;15, ocr_version&＃61;&＃39;PP-OCRv2&＃39;, output&＃61;&＃39;./output/table&＃39;, precision&＃61;&＃39;fp32&＃39;, process_id&＃61;0, rec&＃61;True, rec_algorithm&＃61;&＃39;CRNN&＃39;, rec_batch_num&＃61;6, rec_char_dict_path&＃61;&＃39;/home/lthpc/anaconda3/envs/paddle_env/lib/python3.8/site-packages/paddleocr/ppocr/utils/ppocr_keys_v1.txt&＃39;, rec_image_shape&＃61;&＃39;3, 32, 320&＃39;, rec_model_dir&＃61;&＃39;/home/lthpc/.paddleocr/2.3.0.2/ocr/rec/ch/ch_PP-OCRv2_rec_infer&＃39;, save_log_path&＃61;&＃39;./log_output/&＃39;, show_log&＃61;True, structure_version&＃61;&＃39;STRUCTURE&＃39;, table_char_dict_path&＃61;None, table_char_type&＃61;&＃39;en&＃39;, table_max_len&＃61;488, table_model_dir&＃61;None, total_process_num&＃61;1, type&＃61;&＃39;ocr&＃39;, use_angle_cls&＃61;True, use_dilation&＃61;False, use_gpu&＃61;True, use_mp&＃61;False, use_onnx&＃61;False, use_pdserving&＃61;False, use_space_char&＃61;True, use_tensorrt&＃61;False, vis_font_path&＃61;&＃39;./doc/fonts/simfang.ttf&＃39;, warmup&＃61;True)
[2021/12/08 15:16:00] root INFO: **********./word_1.jpg**********
[2021/12/08 15:16:02] root DEBUG: dt_boxes num : 1, elapse : 2.128263235092163
[2021/12/08 15:16:02] root DEBUG: cls num : 1, elapse : 0.007218360900878906
[2021/12/08 15:16:02] root DEBUG: rec_res num : 1, elapse : 0.01619863510131836
[2021/12/08 15:16:02] root INFO: [[[10.0, 8.0], [344.0, 11.0], [343.0, 66.0], [10.0, 63.0]], (&＃39;韩国小&＃39;, 0.9964866)]


环境安装完成

2.1 检测 &＃43; 方向分类器 &＃43; 识别全流程

from paddleocr import PaddleOCR, draw_ocr
# Paddleocr目前支持中英文、英文、法语、德语、韩语、日语&＃xff0c;可以通过修改lang参数进行切换
# 参数依次为&＃96;ch&＃96;, &＃96;en&＃96;, &＃96;french&＃96;, &＃96;german&＃96;, &＃96;korean&＃96;, &＃96;japan&＃96;。
ocr &＃61; PaddleOCR(use_angle_cls&＃61;True, lang&＃61;"ch") # need to run only once to download and load model into memory
img_path &＃61; &＃39;PaddleOCR/doc/imgs/11.jpg&＃39;
result &＃61; ocr.ocr(img_path, cls&＃61;True)
for line in result:
print(line)
# 显示结果
from PIL import Image
image &＃61; Image.open(img_path).convert(&＃39;RGB&＃39;)
boxes &＃61; [line[0] for line in result]
txts &＃61; [line[1][0] for line in result]
scores &＃61; [line[1][1] for line in result]
im_show &＃61; draw_ocr(image, boxes, txts, scores, font_path&＃61;&＃39;/path/to/PaddleOCR/doc/fonts/simfang.ttf&＃39;)
im_show &＃61; Image.fromarray(im_show)
im_show.save(&＃39;result.jpg&＃39;)


结果是一个list&＃xff0c;每个item包含了文本框&＃xff0c;文字和识别置信度

[[[24.0, 36.0], [304.0, 34.0], [304.0, 72.0], [24.0, 74.0]], [&＃39;纯臻营养护发素&＃39;, 0.964739]]
[[[24.0, 80.0], [172.0, 80.0], [172.0, 104.0], [24.0, 104.0]], [&＃39;产品信息/参数&＃39;, 0.98069626]]
[[[24.0, 109.0], [333.0, 109.0], [333.0, 136.0], [24.0, 136.0]], [&＃39;&＃xff08;45元/每公斤&＃xff0c;100公斤起订&＃xff09;&＃39;, 0.9676722]]
......


结果可视化

 2.2 检测 &＃43; 识别

from paddleocr import PaddleOCR, draw_ocr
ocr &＃61; PaddleOCR() # need to run only once to download and load model into memory
img_path &＃61; &＃39;PaddleOCR/doc/imgs/11.jpg&＃39;
result &＃61; ocr.ocr(img_path, cls&＃61;False)
for line in result:
print(line)
# 显示结果
from PIL import Image
image &＃61; Image.open(img_path).convert(&＃39;RGB&＃39;)
boxes &＃61; [line[0] for line in result]
txts &＃61; [line[1][0] for line in result]
scores &＃61; [line[1][1] for line in result]
im_show &＃61; draw_ocr(image, boxes, txts, scores, font_path&＃61;&＃39;/path/to/PaddleOCR/doc/fonts/simfang.ttf&＃39;)
im_show &＃61; Image.fromarray(im_show)
im_show.save(&＃39;result.jpg&＃39;)


结果是一个list&＃xff0c;每个item包含了文本框&＃xff0c;文字和识别置信度

[[[24.0, 36.0], [304.0, 34.0], [304.0, 72.0], [24.0, 74.0]], [&＃39;纯臻营养护发素&＃39;, 0.964739]]
[[[24.0, 80.0], [172.0, 80.0], [172.0, 104.0], [24.0, 104.0]], [&＃39;产品信息/参数&＃39;, 0.98069626]]
[[[24.0, 109.0], [333.0, 109.0], [333.0, 136.0], [24.0, 136.0]], [&＃39;&＃xff08;45元/每公斤&＃xff0c;100公斤起订&＃xff09;&＃39;, 0.9676722]]
......


结果可视化

2.3 方向分类器 &＃43; 识别

from paddleocr import PaddleOCR
ocr &＃61; PaddleOCR(use_angle_cls&＃61;True) # need to run only once to download and load model into memory
img_path &＃61; &＃39;PaddleOCR/doc/imgs_words/ch/word_1.jpg&＃39;
result &＃61; ocr.ocr(img_path, det&＃61;False, cls&＃61;True)
for line in result:
print(line)


 结果是一个list&＃xff0c;每个item只包含识别结果和识别置信度

[&＃39;韩国小馆&＃39;, 0.9907421]

2.4 单独执行检测

from paddleocr import PaddleOCR, draw_ocr
ocr &＃61; PaddleOCR() # need to run only once to download and load model into memory
img_path &＃61; &＃39;PaddleOCR/doc/imgs/11.jpg&＃39;
result &＃61; ocr.ocr(img_path, rec&＃61;False)
for line in result:
print(line)
# 显示结果
from PIL import Image
image &＃61; Image.open(img_path).convert(&＃39;RGB&＃39;)
im_show &＃61; draw_ocr(image, result, txts&＃61;None, scores&＃61;None, font_path&＃61;&＃39;/path/to/PaddleOCR/doc/fonts/simfang.ttf&＃39;)
im_show &＃61; Image.fromarray(im_show)
im_show.save(&＃39;result.jpg&＃39;)

结果是一个list&＃xff0c;每个item只包含文本框

[[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
[[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
[[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
......

结果可视化

2.5 单独执行识别

from paddleocr import PaddleOCR
ocr &＃61; PaddleOCR() # need to run only once to download and load model into memory
img_path &＃61; &＃39;PaddleOCR/doc/imgs_words/ch/word_1.jpg&＃39;
result &＃61; ocr.ocr(img_path, det&＃61;False)
for line in result:
print(line)


结果是一个list&＃xff0c;每个item只包含识别结果和识别置信度

[&＃39;韩国小馆&＃39;, 0.9907421]

2.6 单独执行方向分类器

from paddleocr import PaddleOCR
ocr &＃61; PaddleOCR(use_angle_cls&＃61;True) # need to run only once to download and load model into memory
img_path &＃61; &＃39;PaddleOCR/doc/imgs_words/ch/word_1.jpg&＃39;
result &＃61; ocr.ocr(img_path, det&＃61;False, rec&＃61;False, cls&＃61;True)
for line in result:
print(line)

结果是一个list&＃xff0c;每个item只包含分类结果和分类置信度

[&＃39;0&＃39;, 0.9999924]
3 替换模型

当内置模型无法满足需求时&＃xff0c;需要使用到自己训练的模型。 首先&＃xff0c;参照 inference.md 第一节转换将检测、分类和识别模型转换为inference模型&＃xff0c;然后按照如下方式使用

from paddleocr import PaddleOCR, draw_ocr
# 模型路径下必须含有model和params文件
ocr &＃61; PaddleOCR(det_model_dir&＃61;&＃39;your_det_model_dir&＃39;, rec_model_dir&＃61;&＃39;your_rec_model_dir&＃39;,
rec_char_dict_path&＃61;&＃39;your_rec_char_dict_path&＃39;, cls_model_dir&＃61;&＃39;your_cls_model_dir&＃39;,
use_angle_cls&＃61;True)
img_path &＃61; &＃39;PaddleOCR/doc/imgs/11.jpg&＃39;
result &＃61; ocr.ocr(img_path, cls&＃61;True)
for line in result:
print(line)
# 显示结果
from PIL import Image
image &＃61; Image.open(img_path).convert(&＃39;RGB&＃39;)
boxes &＃61; [line[0] for line in result]
txts &＃61; [line[1][0] for line in result]
scores &＃61; [line[1][1] for line in result]
im_show &＃61; draw_ocr(image, boxes, txts, scores, font_path&＃61;&＃39;/path/to/PaddleOCR/doc/fonts/simfang.ttf&＃39;)
im_show &＃61; Image.fromarray(im_show)
im_show.save(&＃39;result.jpg&＃39;)

4 参数说明

真香&＃xff0c;后期研究一下如何在自己的数据集上训练模型

•  PaddleOCR/whl.md at release/2.3 · PaddlePaddle/PaddleOCR · GitHub
• https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.3/README_ch.md