diff --git a/__init__.py b/__init__.py index 15a9aca4da19a981b9e678e7cc93e33cf40fc81c..11436094c163db1b91f5ac38f2936a53017016c1 100644 --- a/__init__.py +++ b/__init__.py @@ -16,5 +16,6 @@ from .paddleocr import * __version__ = paddleocr.VERSION __all__ = [ 'PaddleOCR', 'PPStructure', 'draw_ocr', 'draw_structure_result', - 'save_structure_res', 'download_with_progressbar' + 'save_structure_res', 'download_with_progressbar', 'sorted_layout_boxes', + 'convert_info_docx' ] diff --git a/deploy/hubserving/readme.md b/deploy/hubserving/readme.md index 94a570e1895dcb474e51e2b5d2b4dc9a40ee49fd..c583cc96ede437a1f65f9b1bddb69e84b7c54852 100755 --- a/deploy/hubserving/readme.md +++ b/deploy/hubserving/readme.md @@ -20,16 +20,16 @@ PaddleOCR提供2种服务部署方式： # 基于PaddleHub Serving的服务部署 -hubserving服务部署目录下包括文本检测、文本方向分类，文本识别、文本检测+文本方向分类+文本识别3阶段串联，表格识别、PP-Structure和版面分析七种服务包，请根据需求选择相应的服务包进行安装和启动。目录结构如下： +hubserving服务部署目录下包括文本检测、文本方向分类，文本识别、文本检测+文本方向分类+文本识别3阶段串联，版面分析、表格识别和PP-Structure七种服务包，请根据需求选择相应的服务包进行安装和启动。目录结构如下： ``` deploy/hubserving/ └─ ocr_cls 文本方向分类模块服务包 └─ ocr_det 文本检测模块服务包 └─ ocr_rec 文本识别模块服务包 └─ ocr_system 文本检测+文本方向分类+文本识别串联服务包 + └─ structure_layout 版面分析服务包 └─ structure_table 表格识别服务包 └─ structure_system PP-Structure服务包 - └─ structure_layout 版面分析服务包 ``` 每个服务包下包含3个文件。以2阶段串联服务包为例，目录如下： @@ -42,9 +42,9 @@ deploy/hubserving/ocr_system/ ``` ## 1. 近期更新 +* 2022.08.23 新增版面分析服务。 * 2022.05.05 新增PP-OCRv3检测和识别模型。 * 2022.03.30 新增PP-Structure和表格识别两种服务。 -* 2022.08.23 新增版面分析服务。 ## 2. 快速启动服务 以下步骤以检测+识别2阶段串联服务为例，如果只需要检测服务或识别服务，替换相应文件路径即可。 diff --git a/deploy/hubserving/readme_en.md b/deploy/hubserving/readme_en.md index dd0a74e63fe0b138c01d158e577c5c2a0244e94a..f09fe46417c7567305e5ce05a14be74d33450c31 100755 --- a/deploy/hubserving/readme_en.md +++ b/deploy/hubserving/readme_en.md @@ -20,16 +20,16 @@ PaddleOCR provides 2 service deployment methods: # Service deployment based on PaddleHub Serving -The hubserving service deployment directory includes seven service packages: text detection, text angle class, text recognition, text detection+text angle class+text recognition three-stage series connection, table recognition, PP-Structure and layout analysis. Please select the corresponding service package to install and start service according to your needs. The directory is as follows: +The hubserving service deployment directory includes seven service packages: text detection, text angle class, text recognition, text detection+text angle class+text recognition three-stage series connection, layout analysis, table recognition and PP-Structure. Please select the corresponding service package to install and start service according to your needs. The directory is as follows: ``` deploy/hubserving/ └─ ocr_det text detection module service package └─ ocr_cls text angle class module service package └─ ocr_rec text recognition module service package └─ ocr_system text detection+text angle class+text recognition three-stage series connection service package + └─ structure_layout layout analysis service package └─ structure_table table recognition service package └─ structure_system PP-Structure service package - └─ structure_layout layout analysis service package ``` Each service pack contains 3 files. Take the 2-stage series connection service package as an example, the directory is as follows: diff --git a/paddleocr.py b/paddleocr.py index 2adddfbe8293a29cc1f292755f70f0fa03f82c1d..bada383612725608d19596ea6f331758a8aba53c 100644 --- a/paddleocr.py +++ b/paddleocr.py @@ -562,7 +562,7 @@ class PPStructure(StructureSystem): params.table_model_dir, os.path.join(BASE_DIR, 'whl', 'table'), table_model_config['url']) layout_model_config = get_model_config( - 'STRUCTURE', params.structure_version, 'layout', 'ch') + 'STRUCTURE', params.structure_version, 'layout', lang) params.layout_model_dir, layout_url = confirm_model_dir_url( params.layout_model_dir, os.path.join(BASE_DIR, 'whl', 'layout'), layout_model_config['url']) @@ -584,7 +584,7 @@ class PPStructure(StructureSystem): logger.debug(params) super().__init__(params) - def __call__(self, img, return_ocr_result_in_table=False): + def __call__(self, img, return_ocr_result_in_table=False, img_idx=0): if isinstance(img, str): # download net image if img.startswith('http'): @@ -602,7 +602,8 @@ class PPStructure(StructureSystem): if isinstance(img, np.ndarray) and len(img.shape) == 2: img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) - res, _ = super().__call__(img, return_ocr_result_in_table) + res, _ = super().__call__( + img, return_ocr_result_in_table, img_idx=img_idx) return res @@ -637,25 +638,54 @@ def main(): for line in result: logger.info(line) elif args.type == 'structure': - result = engine(img_path) - save_structure_res(result, args.output, img_name) + img, flag_gif, flag_pdf = check_and_read(img_path) + if not flag_gif and not flag_pdf: + img = cv2.imread(img_path) - if args.recovery: - try: - from ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx - img = cv2.imread(img_path) + if not flag_pdf: + if img is None: + logger.error("error in loading image:{}".format(image_file)) + continue + img_paths = [[img_path, img]] + else: + img_paths = [] + for index, pdf_img in enumerate(img): + os.makedirs( + os.path.join(args.output, img_name), exist_ok=True) + pdf_img_path = os.path.join(args.output, img_name, img_name + + '_' + str(index) + '.jpg') + cv2.imwrite(pdf_img_path, pdf_img) + img_paths.append([pdf_img_path, pdf_img]) + + all_res = [] + for index, (new_img_path, img) in enumerate(img_paths): + logger.info('processing {}/{} page:'.format(index + 1, + len(img_paths))) + new_img_name = os.path.basename(new_img_path).split('.')[0] + result = engine(new_img_path, img_idx=index) + save_structure_res(result, args.output, img_name, index) + + if args.recovery and result != []: + from copy import deepcopy + from ppstructure.recovery.recovery_to_doc import sorted_layout_boxes h, w, _ = img.shape - res = sorted_layout_boxes(result, w) - convert_info_docx(img, res, args.output, img_name, + result_cp = deepcopy(result) + result_sorted = sorted_layout_boxes(result_cp, w) + all_res += result_sorted + + for item in result: + item.pop('img') + item.pop('res') + logger.info(item) + logger.info('result save to {}'.format(args.output)) + + if args.recovery and all_res != []: + try: + from ppstructure.recovery.recovery_to_doc import convert_info_docx + convert_info_docx(img, all_res, args.output, img_name, args.save_pdf) except Exception as ex: logger.error( "error in layout recovery image:{}, err msg: {}".format( img_name, ex)) continue - - for item in result: - item.pop('img') - item.pop('res') - logger.info(item) - logger.info('result save to {}'.format(args.output)) diff --git a/ppstructure/docs/quickstart.md b/ppstructure/docs/quickstart.md index 52331abef78c27437ecb625b77d466bd1f62bfb5..b9367cab327a2f6232e34431c12532db03c75389 100644 --- a/ppstructure/docs/quickstart.md +++ b/ppstructure/docs/quickstart.md @@ -102,6 +102,8 @@ paddleocr --image_dir=ppstructure/docs/table/table.jpg --type=structure --layout paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true # 英文测试图 paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en' +# pdf测试文件 +paddleocr --image_dir=ppstructure/recovery/UnrealText.pdf --type=structure --recovery=true --lang='en' ``` diff --git a/ppstructure/docs/quickstart_en.md b/ppstructure/docs/quickstart_en.md index 83a852a33519cfaef7f5da7929fd3e33140b5c19..52f4e34f8fd500b2de62f63246262890764b6e9a 100644 --- a/ppstructure/docs/quickstart_en.md +++ b/ppstructure/docs/quickstart_en.md @@ -85,6 +85,8 @@ Please refer to: [Key Information Extraction](../kie/README.md) . paddleocr --image_dir=PaddleOCR/ppstructure/docs/table/1.png --type=structure --recovery=true # English pic paddleocr --image_dir=PaddleOCR/ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en' +# pdf file +paddleocr --image_dir=ppstructure/recovery/UnrealText.pdf --type=structure --recovery=true --lang='en' ``` diff --git a/ppstructure/docs/recovery/UnrealText.pdf b/ppstructure/docs/recovery/UnrealText.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0b5cf961af4ebf09cb96fc3f09fb9c19abec68f1 Binary files /dev/null and b/ppstructure/docs/recovery/UnrealText.pdf differ diff --git a/ppstructure/docs/recovery/recovery_ch.jpg b/ppstructure/docs/recovery/recovery_ch.jpg new file mode 100644 index 0000000000000000000000000000000000000000..df5a5063f036053673041b92a01f288b3e1d246b Binary files /dev/null and b/ppstructure/docs/recovery/recovery_ch.jpg differ diff --git a/ppstructure/layout/README_ch.md b/ppstructure/layout/README_ch.md index d5598fc1a896ea4cfcc94619e1744b9b7ec288b3..f8d1978e25d7fb17cfd3fcb363b4ce981e19c8dc 100644 --- a/ppstructure/layout/README_ch.md +++ b/ppstructure/layout/README_ch.md @@ -160,11 +160,13 @@ json文件包含所有图像的标注，数据以字典嵌套的方式存放， ``` mkdir pretrained_model cd pretrained_model -# 下载PubLayNet预训练模型 -wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_layout.pdparams +# 下载PubLayNet预训练模型（直接体验模型评估、预测、动转静） +wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams +# 下载PubLaynet推理模型（直接体验模型推理） +wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar ``` -下载更多[版面分析模型](../docs/models_list.md)（中文CDLA数据集预训练模型、表格预训练模型） +如果测试图片为中文，可以下载中文CDLA数据集的预训练模型，识别10类文档区域：Table、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation，在[版面分析模型](../docs/models_list.md)中下载`picodet_lcnet_x1_0_fgd_layout_cdla`模型的训练模型和推理模型。如果只检测图片中的表格区域，可以下载表格数据集的预训练模型，在[版面分析模型](../docs/models_list.md)中下载`picodet_lcnet_x1_0_fgd_layout_table`模型的训练模型和推理模型。 ### 4.1. 启动训练 @@ -216,14 +218,14 @@ TestDataset: # 单卡训练 export CUDA_VISIBLE_DEVICES=0 python3 tools/train.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - --eval + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + --eval # 多卡训练，通过--gpus参数指定卡号 export CUDA_VISIBLE_DEVICES=0,1,2,3 python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - --eval + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + --eval ``` **注意：**如果训练时显存out memory，将TrainReader中batch_size调小，同时LearningRate中base_lr等比例减小。发布的config均由8卡训练得到，如果改变GPU卡数为1，那么base_lr需要减小8倍。 @@ -252,9 +254,9 @@ PaddleDetection支持了基于FGD([Focal and Global Knowledge Distillation for D # 单卡训练 export CUDA_VISIBLE_DEVICES=0 python3 tools/train.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ - --eval + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ + --eval ``` - `-c`: 指定模型配置文件。 @@ -269,8 +271,8 @@ python3 tools/train.py \ ```bash # GPU 评估， weights 为待测权重 python3 tools/eval.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - -o weights=./output/picodet_lcnet_x1_0_layout/best_model + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + -o weights=./output/picodet_lcnet_x1_0_layout/best_model ``` 会输出以下信息，打印出mAP、AP0.5等信息。 @@ -292,13 +294,13 @@ python3 tools/eval.py \ [08/15 07:07:09] ppdet.engine INFO: Best test bbox ap is 0.935. ``` -使用FGD蒸馏模型进行评估： +若使用**提供的预训练模型进行评估**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，执行如下命令进行评估： ``` python3 tools/eval.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ - -o weights=output/picodet_lcnet_x2_5_layout/best_model + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ + -o weights=output/picodet_lcnet_x2_5_layout/best_model ``` - `-c`: 指定模型配置文件。 @@ -325,18 +327,16 @@ python3 tools/infer.py \ - `--output_dir`: 指定可视化结果保存路径。 - `--draw_threshold`:指定绘制结果框的NMS阈值。 -预测图片如下所示，图片会存储在`output_dir`路径中。 - -使用FGD蒸馏模型进行测试： +若使用**提供的预训练模型进行预测**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，执行如下命令进行预测： ``` python3 tools/infer.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ - -o weights='output/picodet_lcnet_x2_5_layout/best_model.pdparams' \ - --infer_img='docs/images/layout.jpg' \ - --output_dir=output_dir/ \ - --draw_threshold=0.5 + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ + -o weights='output/picodet_lcnet_x2_5_layout/best_model.pdparams' \ + --infer_img='docs/images/layout.jpg' \ + --output_dir=output_dir/ \ + --draw_threshold=0.5 ``` @@ -351,9 +351,9 @@ inference 模型（`paddle.jit.save`保存的模型） 一般是模型训练， ```bash python3 tools/export_model.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - -o weights=output/picodet_lcnet_x1_0_layout/best_model \ - --output_dir=output_inference/ + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + -o weights=output/picodet_lcnet_x1_0_layout/best_model \ + --output_dir=output_inference/ ``` * 如无需导出后处理，请指定：`-o export.benchmark=True`（如果-o已出现过，此处删掉-o） @@ -368,27 +368,27 @@ output_inference/picodet_lcnet_x1_0_layout/ └── model.pdmodel # inference模型的模型结构文件 ``` -FGD蒸馏模型转inference模型步骤如下： +若使用**提供的预训练模型转Inference模型**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，模型转inference模型步骤如下： ```bash python3 tools/export_model.py \ - -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ - --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ - -o weights=./output/picodet_lcnet_x2_5_layout/best_model \ - --output_dir=output_inference/ + -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \ + --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \ + -o weights=./output/picodet_lcnet_x2_5_layout/best_model \ + --output_dir=output_inference/ ``` ### 6.2 模型推理 -版面恢复任务进行推理，可以执行如下命令： +若使用**提供的推理训练模型推理**，或使用**FGD蒸馏训练的模型**，更换`model_dir`推理模型路径，执行如下命令进行推理： ```bash python3 deploy/python/infer.py \ - --model_dir=output_inference/picodet_lcnet_x1_0_layout/ \ - --image_file=docs/images/layout.jpg \ - --device=CPU + --model_dir=output_inference/picodet_lcnet_x1_0_layout/ \ + --image_file=docs/images/layout.jpg \ + --device=CPU ``` - --device：指定GPU、CPU设备 diff --git a/ppstructure/predict_system.py b/ppstructure/predict_system.py index da6b0770bba3ca8e01476ed433b7a48a85ba887c..71147d3af8ec666d368234270dcb0d16aaf91938 100644 --- a/ppstructure/predict_system.py +++ b/ppstructure/predict_system.py @@ -227,65 +227,39 @@ def main(args): if img is None: logger.error("error in loading image:{}".format(image_file)) continue - res, time_dict = structure_sys(img) + imgs = [img] + else: + imgs = img - if structure_sys.mode == 'structure': - save_structure_res(res, save_folder, img_name) + all_res = [] + for index, img in enumerate(imgs): + res, time_dict = structure_sys(img, img_idx=index) + if structure_sys.mode == 'structure' and res != []: + save_structure_res(res, save_folder, img_name, index) draw_img = draw_structure_result(img, res, args.vis_font_path) - img_save_path = os.path.join(save_folder, img_name, 'show.jpg') + img_save_path = os.path.join(save_folder, img_name, + 'show_{}.jpg'.format(index)) elif structure_sys.mode == 'kie': raise NotImplementedError # draw_img = draw_ser_results(img, res, args.vis_font_path) # img_save_path = os.path.join(save_folder, img_name + '.jpg') - cv2.imwrite(img_save_path, draw_img) - logger.info('result save to {}'.format(img_save_path)) - if args.recovery: - try: - from ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx - h, w, _ = img.shape - res = sorted_layout_boxes(res, w) - convert_info_docx(img, res, save_folder, img_name, - args.save_pdf) - except Exception as ex: - logger.error( - "error in layout recovery image:{}, err msg: {}".format( - image_file, ex)) - continue - else: - pdf_imgs = img - all_res = [] - for index, img in enumerate(pdf_imgs): - - res, time_dict = structure_sys(img, index) - if structure_sys.mode == 'structure' and res != []: - save_structure_res(res, save_folder, img_name, index) - draw_img = draw_structure_result(img, res, - args.vis_font_path) - img_save_path = os.path.join(save_folder, img_name, - 'show_{}.jpg'.format(index)) - elif structure_sys.mode == 'kie': - raise NotImplementedError - # draw_img = draw_ser_results(img, res, args.vis_font_path) - # img_save_path = os.path.join(save_folder, img_name + '.jpg') - if res != []: - cv2.imwrite(img_save_path, draw_img) - logger.info('result save to {}'.format(img_save_path)) - if args.recovery and res != []: - from ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx - h, w, _ = img.shape - res = sorted_layout_boxes(res, w) - all_res += res - - if args.recovery and all_res != []: - try: - convert_info_docx(img, all_res, save_folder, img_name, - args.save_pdf) - except Exception as ex: - logger.error( - "error in layout recovery image:{}, err msg: {}".format( - image_file, ex)) - continue + if res != []: + cv2.imwrite(img_save_path, draw_img) + logger.info('result save to {}'.format(img_save_path)) + if args.recovery and res != []: + from ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx + h, w, _ = img.shape + res = sorted_layout_boxes(res, w) + all_res += res + if args.recovery and all_res != []: + try: + convert_info_docx(img, all_res, save_folder, img_name, + args.save_pdf) + except Exception as ex: + logger.error("error in layout recovery image:{}, err msg: {}". + format(image_file, ex)) + continue logger.info("Predict time : {:.3f}s".format(time_dict['all'])) diff --git a/ppstructure/recovery/README.md b/ppstructure/recovery/README.md index 90a6a2c3c4189dc885d698e4cac2d1a24a49d1df..59aef707dd67799bb46dc18dc58f883c502c8b86 100644 --- a/ppstructure/recovery/README.md +++ b/ppstructure/recovery/README.md @@ -8,6 +8,7 @@ English | [简体中文](README_ch.md) - [3. Quick Start](#3) - [3.1 Download models](#3.1) - [3.2 Layout recovery](#3.2) + - [4. More](#4) @@ -15,13 +16,16 @@ English | [简体中文](README_ch.md) Layout recovery means that after OCR recognition, the content is still arranged like the original document pictures, and the paragraphs are output to word document in the same order. -Layout recovery combines [layout analysis](../layout/README.md)、[table recognition](../table/README.md) to better recover images, tables, titles, etc. -The following figure shows the result： +Layout recovery combines [layout analysis](../layout/README.md)、[table recognition](../table/README.md) to better recover images, tables, titles, etc. supports input files in PDF and document image formats in Chinese and English. The following figure shows the effect of restoring the layout of English and Chinese documents:
+
+ +
+ ## 2. Install @@ -35,7 +39,7 @@ The following figure shows the result： ```bash python3 -m pip install --upgrade pip -# GPU installation +# If you have cuda9 or cuda10 installed on your machine, please run the following command to install python3 -m pip install "paddlepaddle-gpu" -i https://mirror.baidu.com/pypi/simple # CPU installation @@ -62,6 +66,8 @@ git clone https://gitee.com/paddlepaddle/PaddleOCR - **(2) Install recovery's `requirements`** +The layout restoration is exported as docx and PDF files, so python-docx and docx2pdf API need to be installed, and fitz and PyMuPDF apis need to be installed to process the input files in pdf format. + ```bash python3 -m pip install -r ppstructure/recovery/requirements.txt ```` @@ -70,6 +76,16 @@ python3 -m pip install -r ppstructure/recovery/requirements.txt ## 3. Quick Start +Through layout analysis, we divided the image/PDF documents into regions, located the key regions, such as text, table, picture, etc., and recorded the location, category, and regional pixel value information of each region. Different regions are processed separately, where: + +- OCR detection and recognition is performed in the text area, and the coordinates of the OCR detection box and the text content information are added on the basis of the previous information + +- The table area identifies tables and records html and text information of tables +- Save the image directly + +We can restore the test picture through the layout information, OCR detection and recognition structure, table information, and saved pictures. + + ### 3.1 Download models @@ -85,9 +101,11 @@ https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && ta # Download the recognition model of the ultra-lightweight English PP-OCRv3 model and unzip it wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar && tar xf en_PP-OCRv3_rec_infer.tar # Download the ultra-lightweight English table inch model and unzip it -wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar +wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_infer.tar +tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar # Download the layout model of publaynet dataset and unzip it -wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar && tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar +wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar +tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar cd .. ``` If input is Chinese document，download Chinese models: @@ -128,3 +146,15 @@ Field： - recovery：whether to enable layout of recovery, default False - save_pdf：when recovery file, whether to save pdf file, default False - output：save the recovery result path + + + +## 4. More + +For training, evaluation and inference tutorial for text detection models, please refer to [text detection doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/detection.md). + +For training, evaluation and inference tutorial for text recognition models, please refer to [text recognition doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/recognition.md). + +For training, evaluation and inference tutorial for layout analysis models, please refer to [layout analysis doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/layout/README_ch.md) + +For training, evaluation and inference tutorial for table recognition models, please refer to [table recognition doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/table/README_ch.md) diff --git a/ppstructure/recovery/README_ch.md b/ppstructure/recovery/README_ch.md index 9215976d37e89c7f02a61a5dfcf2127ff98c998e..ae3b7ed82464f513af585542ef8e92d66f2c8756 100644 --- a/ppstructure/recovery/README_ch.md +++ b/ppstructure/recovery/README_ch.md @@ -10,6 +10,7 @@ - [3. 使用](#3) - [3.1 下载模型](#3.1) - [3.2 版面恢复](#3.2) +- [4. 更多](#4) @@ -18,11 +19,14 @@ 版面恢复就是在OCR识别后，内容仍然像原文档图片那样排列着，段落不变、顺序不变的输出到word文档中等。 -版面恢复结合了[版面分析](../layout/README_ch.md)、[表格识别](../table/README_ch.md)技术，从而更好地恢复图片、表格、标题等内容，支持pdf文档、文档图片格式的输入文件，下图展示了版面恢复的结果： +版面恢复结合了[版面分析](../layout/README_ch.md)、[表格识别](../table/README_ch.md)技术，从而更好地恢复图片、表格、标题等内容，支持中、英文pdf文档、文档图片格式的输入文件，下图分别展示了英文文档和中文文档版面恢复的效果：
+
+ +
@@ -37,10 +41,10 @@ ```bash python3 -m pip install --upgrade pip -# GPU安装 +# 您的机器安装的是CUDA9或CUDA10，请运行以下命令安装 python3 -m pip install "paddlepaddle-gpu" -i https://mirror.baidu.com/pypi/simple -# CPU安装 +# 您的机器是CPU，请运行以下命令安装 python3 -m pip install "paddlepaddle" -i https://mirror.baidu.com/pypi/simple ``` @@ -64,6 +68,8 @@ git clone https://gitee.com/paddlepaddle/PaddleOCR - **（2）安装recovery的`requirements`** +版面恢复导出为docx、pdf文件，所以需要安装python-docx、docx2pdf API，同时处理pdf格式的输入文件，需要安装fitz、PyMuPDF API。 + ```bash python3 -m pip install -r ppstructure/recovery/requirements.txt ``` @@ -72,11 +78,20 @@ python3 -m pip install -r ppstructure/recovery/requirements.txt ## 3. 使用 +我们通过版面分析对图片/pdf形式的文档进行区域划分，定位其中的关键区域，如文字、表格、图片等，记录每个区域的位置、类别、区域像素值信息。对不同的区域分别处理，其中： + +- 文字区域直接进行OCR检测和识别，在之前信息基础上增加OCR检测框坐标和文本内容信息 + +- 表格区域进行表格识别，记录表格html和文字信息 +- 图片直接保存 + +我们通过版面信息、OCR检测和识别结构、表格信息、保存的图片，对测试图片进行恢复即可。 + ### 3.1 下载模型 -如果输入为英文文档类型，下载英文模型 +如果输入为英文文档类型，下载OCR检测和识别、版面分析、表格识别的英文模型 ```bash cd PaddleOCR/ppstructure @@ -88,9 +103,11 @@ wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar # 下载英文超轻量PP-OCRv3识别模型并解压 wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar && tar xf en_PP-OCRv3_rec_infer.tar # 下载英文表格识别模型并解压 -wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar +wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_infer.tar +tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar # 下载英文版面分析模型 -wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar && tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar +wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar +tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar cd .. ``` @@ -135,3 +152,15 @@ python3 predict_system.py \ - recovery：是否进行版面恢复，默认False - save_pdf：进行版面恢复导出docx文档的同时，是否保存为pdf文件，默认为False - output：版面恢复结果保存路径 + + + +## 4. 更多 + +关于OCR检测模型的训练评估与推理，请参考：[文本检测教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/detection.md) + +关于OCR识别模型的训练评估与推理，请参考：[文本识别教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/recognition.md) + +关于版面分析模型的训练评估与推理，请参考：[版面分析教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/layout/README_ch.md) + +关于表格识别模型的训练评估与推理，请参考：[表格识别教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/table/README_ch.md) diff --git a/ppstructure/recovery/requirements.txt b/ppstructure/recovery/requirements.txt index 37e6151bdcd546d032f86e256d00d566c1b6b488..b118a41e516ec20e5807030649943e5f7d848107 100644 --- a/ppstructure/recovery/requirements.txt +++ b/ppstructure/recovery/requirements.txt @@ -1,5 +1,5 @@ python-docx docx2pdf fitz -PyMuPDF +PyMuPDF==1.16.14 beautifulsoup4 \ No newline at end of file