diff --git a/README.md b/README.md index 4d6fb7fb869c4ac7fc0ac03d266b9790cd36dfbc..ce8f7cbcbb19387950c2d6e8a2538258dbcc19bc 100644 --- a/README.md +++ b/README.md @@ -1,2 +1,149 @@ -# PaddleOCR -OCR algorithms with PaddlePaddle （still under develop) + +# 简介 +PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力使用者训练出更好的模型，并应用落地。 + +## 特性： +- 超轻量级模型 + - (检测模型4.1M + 识别模型4.5M = 8.6M) +- 支持竖排文字识别 + - (单模型同时支持横排和竖排文字识别) +- 支持长文本识别 +- 支持中英文数字组合识别 +- 提供训练代码 +- 支持模型部署 + + +## 文档教程 +- [快速安装](./doc/installation.md) +- [文本识别模型训练/评估/预测](./doc/detection.md) +- [文本预测模型训练/评估/预测](./doc/recognition.md) +- [基于inference model预测](./doc/) + +### **快速开始** + +下载inference模型 +``` +# 创建inference模型保存目录 +mkdir inference && cd inference && mkdir det && mkdir rec +# 下载检测inference模型/ 识别 inference 模型 +wget -P ./inference https://paddleocr.bj.bcebos.com/inference.tar +``` + +实现文本检测、识别串联推理，预测$image_dir$指定的单张图像： +``` +export PYTHONPATH=. +python tools/infer/predict_eval.py --image_dir="/Demo.jpg" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" +``` +在执行预测时，通过参数det_model_dir以及rec_model_dir设置存储inference 模型的路径。 + +实现文本检测、识别串联推理，预测$image_dir$指指定文件夹下的所有图像： +``` +python tools/infer/predict_eval.py --image_dir="/test_imgs/" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" +``` + + + +## 文本检测算法: + +PaddleOCR开源的文本检测算法列表： +- [x] [EAST](https://arxiv.org/abs/1704.03155) +- [x] [DB](https://arxiv.org/abs/1911.08947) +- [ ] [SAST](https://arxiv.org/abs/1908.05498) + + +算法效果： +|模型|骨干网络|Hmean| +|-|-|-| +|EAST|[ResNet50_vd](https://paddleocr.bj.bcebos.com/det_r50_vd_east.tar)|85.85%| +|EAST|[MobileNetV3](https://paddleocr.bj.bcebos.com/det_mv3_east.tar)|79.08%| +|DB|[ResNet50_vd](https://paddleocr.bj.bcebos.com/det_r50_vd_db.tar)|83.30%| +|DB|[MobileNetV3](https://paddleocr.bj.bcebos.com/det_mv3_db.tar)|73.00%| + +PaddleOCR文本检测算法的训练与使用请参考[文档](./doc/detection.md)。 + +## 文本识别算法: + +PaddleOCR开源的文本识别算法列表： +- [x] [CRNN](https://arxiv.org/abs/1507.05717) +- [x] [DTRB](https://arxiv.org/abs/1904.01906) +- [ ] [Rosetta](https://arxiv.org/abs/1910.05085) +- [ ] [STAR-Net](http://www.bmva.org/bmvc/2016/papers/paper043/index.html) +- [ ] [RARE](https://arxiv.org/abs/1603.03915v1) +- [ ] [SRN]((https://arxiv.org/abs/2003.12294))(百度自研) + +算法效果如下表所示，精度指标是在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上的评测结果的平均值。 + +|模型|骨干网络|ACC| +|-|-|-| +|Rosetta|[Resnet34_vd](https://paddleocr.bj.bcebos.com/rec_r34_vd_none_none_ctc.tar)|80.24%| +|Rosetta|[MobileNetV3](https://paddleocr.bj.bcebos.com/rec_mv3_none_none_ctc.tar)|78.16%| +|CRNN|[Resnet34_vd](https://paddleocr.bj.bcebos.com/rec_r34_vd_none_bilstm_ctc.tar)|82.20%| +|CRNN|[MobileNetV3](https://paddleocr.bj.bcebos.com/rec_mv3_none_bilstm_ctc.tar)|79.37%| +|STAR-Net|[Resnet34_vd](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_ctc.tar)|83.93%| +|STAR-Net|[MobileNetV3](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_ctc.tar)|81.56%| +|RARE|[Resnet34_vd](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_attn.tar)|84.90%| +|RARE|[MobileNetV3](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_attn.tar)|83.32%| + +PaddleOCR文本识别算法的训练与使用请参考[文档](./doc/recognition.md)。 + +## TODO +**端到端OCR算法** +PaddleOCR即将开源百度自研端对端OCR模型[End2End-PSL](https://arxiv.org/abs/1909.07808)，敬请关注。 +- [ ] End2End-PSL (comming soon) + + + +# 参考文献 +``` +1. EAST: +@inproceedings{zhou2017east, + title={EAST: an efficient and accurate scene text detector}, + author={Zhou, Xinyu and Yao, Cong and Wen, He and Wang, Yuzhi and Zhou, Shuchang and He, Weiran and Liang, Jiajun}, + booktitle={Proceedings of the IEEE conference on Computer Vision and Pattern Recognition}, + pages={5551--5560}, + year={2017} +} + +2. DB: +@article{liao2019real, + title={Real-time Scene Text Detection with Differentiable Binarization}, + author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang}, + journal={arXiv preprint arXiv:1911.08947}, + year={2019} +} + +3. DTRB: +@inproceedings{baek2019wrong, + title={What is wrong with scene text recognition model comparisons? dataset and model analysis}, + author={Baek, Jeonghun and Kim, Geewook and Lee, Junyeop and Park, Sungrae and Han, Dongyoon and Yun, Sangdoo and Oh, Seong Joon and Lee, Hwalsuk}, + booktitle={Proceedings of the IEEE International Conference on Computer Vision}, + pages={4715--4723}, + year={2019} +} + +4. SAST: +@inproceedings{wang2019single, + title={A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning}, + author={Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming}, + booktitle={Proceedings of the 27th ACM International Conference on Multimedia}, + pages={1277--1285}, + year={2019} +} + +5. SRN: +@article{yu2020towards, + title={Towards Accurate Scene Text Recognition with Semantic Reasoning Networks}, + author={Yu, Deli and Li, Xuan and Zhang, Chengquan and Han, Junyu and Liu, Jingtuo and Ding, Errui}, + journal={arXiv preprint arXiv:2003.12294}, + year={2020} +} + +6. end2end-psl: +@inproceedings{sun2019chinese, + title={Chinese Street View Text: Large-scale Chinese Text Reading with Partially Supervised Learning}, + author={Sun, Yipeng and Liu, Jiaming and Liu, Wei and Han, Junyu and Ding, Errui and Liu, Jingtuo}, + booktitle={Proceedings of the IEEE International Conference on Computer Vision}, + pages={9086--9095}, + year={2019} +} +``` diff --git a/configs/rec/rec_chinese_lite_train.yml b/configs/rec/rec_chinese_lite_train.yml new file mode 100755 index 0000000000000000000000000000000000000000..279aa6999a727045e50ce079e9a33bbdda321a89 --- /dev/null +++ b/configs/rec/rec_chinese_lite_train.yml @@ -0,0 +1,42 @@ +Global: + algorithm: CRNN + use_gpu: true + epoch_num: 3000 + log_smooth_window: 20 + print_batch_step: 10 + save_model_dir: ./output/rec_CRNN + save_epoch_step: 3 + eval_batch_step: 2000 + train_batch_size_per_card: 256 + test_batch_size_per_card: 256 + image_shape: [3, 32, 100] + max_text_length: 25 + character_type: ch + character_dict_path: ./ppocr/utils/ppocr_keys_v1.txt + loss_type: ctc + reader_yml: ./configs/rec/rec_chinese_reader.yml + pretrain_weights: ./pretrain_models/CRNN/best_accuracy + checkpoints: + save_inference_dir: +Architecture: + function: ppocr.modeling.architectures.rec_model,RecModel + +Backbone: + function: ppocr.modeling.backbones.rec_mobilenet_v3,MobileNetV3 + scale: 0.5 + model_name: small + +Head: + function: ppocr.modeling.heads.rec_ctc_head,CTCPredict + encoder_type: rnn + SeqRNN: + hidden_size: 48 + +Loss: + function: ppocr.modeling.losses.rec_ctc_loss,CTCLoss + +Optimizer: + function: ppocr.optimizer,AdamDecay + base_lr: 0.0005 + beta1: 0.9 + beta2: 0.999 diff --git a/configs/rec/rec_chinese_reader.yml b/configs/rec/rec_chinese_reader.yml new file mode 100755 index 0000000000000000000000000000000000000000..f09a1ea72e6d929d0446fbbf51ca218e52ae5b3e --- /dev/null +++ b/configs/rec/rec_chinese_reader.yml @@ -0,0 +1,14 @@ +TrainReader: + reader_function: ppocr.data.rec.dataset_traversal,SimpleReader + num_workers: 8 + img_set_dir: ./train_data + label_file_path: ./train_data/rec_gt_train.txt + +EvalReader: + reader_function: ppocr.data.rec.dataset_traversal,SimpleReader + img_set_dir: ./train_data + label_file_path: ./train_data/rec_gt_test.txt + +TestReader: + reader_function: ppocr.data.rec.dataset_traversal,SimpleReader + infer_img: ./infer_img diff --git a/configs/rec/rec_icdar15_train.yml b/configs/rec/rec_icdar15_train.yml index 35aba6684eecb6a932b74dd7e1ebd32c9f0f2f43..dd6796af60243ce506945167c7621214ced09352 100755 --- a/configs/rec/rec_icdar15_train.yml +++ b/configs/rec/rec_icdar15_train.yml @@ -11,8 +11,8 @@ Global: test_batch_size_per_card: 256 image_shape: [3, 32, 100] max_text_length: 25 - character_type: ch - character_dict_path: ./ppocr/utils/ic15_dict.txt + character_type: en + character_dict_path: /workspace/PaddleOCR/train_data/ic15_dict.txt loss_type: ctc reader_yml: ./configs/rec/rec_icdar15_reader.yml pretrain_weights: ./pretrain_models/CRNN/best_accuracy @@ -24,13 +24,13 @@ Architecture: Backbone: function: ppocr.modeling.backbones.rec_mobilenet_v3,MobileNetV3 scale: 0.5 - model_name: small + model_name: large Head: function: ppocr.modeling.heads.rec_ctc_head,CTCPredict encoder_type: rnn SeqRNN: - hidden_size: 48 + hidden_size: 96 Loss: function: ppocr.modeling.losses.rec_ctc_loss,CTCLoss diff --git a/doc/README.md b/doc/README.md deleted file mode 100644 index 6a9adfa608aacbaa897a3625ebd6295c6234f860..0000000000000000000000000000000000000000 --- a/doc/README.md +++ /dev/null @@ -1,151 +0,0 @@ - -# 简介 -PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力使用者训练出更好的模型，并应用落地。 - -## 特性： -- 超轻量级模型 - - (检测模型4.1M + 识别模型4.5M = 8.6M) -- 支持竖排文字识别 - - (单模型同时支持横排和竖排文字识别) -- 支持长文本识别 -- 支持中英文数字组合识别 -- 提供训练代码 -- 支持模型部署 - - -## 文档教程 -- [快速安装](./doc/installation.md) -- [文本识别模型训练/评估/预测](./doc/detection.md) -- [文本预测模型训练/评估/预测](./doc/recognition.md) -- [基于inference model预测](./doc/) - -### **快速开始** - -下载inference模型 -``` -# 创建inference模型保存目录 -mkdir inference && cd inference && mkdir det && mkdir rec -# 下载检测inference模型 -wget -P ./inference/det 检测inference模型链接 -# 下载识别inference模型 -wget -P ./inferencee/rec 识别inference模型链接 -``` - -实现文本检测、识别串联推理，预测$image_dir$指定的单张图像： -``` -export PYTHONPATH=. -python tools/infer/predict_eval.py --image_dir="/Demo.jpg" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" -``` -在执行预测时，通过参数det_model_dir以及rec_model_dir设置存储inference 模型的路径。 - -实现文本检测、识别串联推理，预测$image_dir$指指定文件夹下的所有图像： -``` -python tools/infer/predict_eval.py --image_dir="/test_imgs/" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" -``` - - - -## 文本检测算法: - -PaddleOCR开源的文本检测算法列表： -- [x] [EAST](https://arxiv.org/abs/1704.03155) -- [x] [DB](https://arxiv.org/abs/1911.08947) -- [ ] [SAST](https://arxiv.org/abs/1908.05498) - - -算法效果： -|模型|骨干网络|Hmean| -|-|-|-| -|EAST|ResNet50_vd|85.85%| -|EAST|MobileNetV3|79.08%| -|DB|ResNet50_vd|83.30%| -|DB|MobileNetV3|73.00%| - -PaddleOCR文本检测算法的训练与使用请参考[文档](./doc/detection.md)。 - -## 文本识别算法: - -PaddleOCR开源的文本识别算法列表： -- [x] [CRNN](https://arxiv.org/abs/1507.05717) -- [x] [DTRB](https://arxiv.org/abs/1904.01906) -- [ ] [Rosetta](https://arxiv.org/abs/1910.05085) -- [ ] [STAR-Net](http://www.bmva.org/bmvc/2016/papers/paper043/index.html) -- [ ] [RARE](https://arxiv.org/abs/1603.03915v1) -- [ ] [SRN]((https://arxiv.org/abs/2003.12294))(百度自研) - -算法效果如下表所示，精度指标是在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上的评测结果的平均值。 - -|模型|骨干网络|ACC| -|-|-|-| -|Rosetta|Resnet34_vd|80.24%| -|Rosetta|MobileNetV3|78.16%| -|CRNN|Resnet34_vd|82.20%| -|CRNN|MobileNetV3|79.37%| -|STAR-Net|Resnet34_vd|83.93%| -|STAR-Net|MobileNetV3|81.56%| -|RARE|Resnet34_vd|84.90%| -|RARE|MobileNetV3|83.32%| - -PaddleOCR文本识别算法的训练与使用请参考[文档](./doc/recognition.md)。 - -## TODO -**端到端OCR算法** -PaddleOCR即将开源百度自研端对端OCR模型[End2End-PSL](https://arxiv.org/abs/1909.07808)，敬请关注。 -- [ ] End2End-PSL (comming soon) - - - -# 参考文献 -``` -1. EAST: -@inproceedings{zhou2017east, - title={EAST: an efficient and accurate scene text detector}, - author={Zhou, Xinyu and Yao, Cong and Wen, He and Wang, Yuzhi and Zhou, Shuchang and He, Weiran and Liang, Jiajun}, - booktitle={Proceedings of the IEEE conference on Computer Vision and Pattern Recognition}, - pages={5551--5560}, - year={2017} -} - -2. DB: -@article{liao2019real, - title={Real-time Scene Text Detection with Differentiable Binarization}, - author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang}, - journal={arXiv preprint arXiv:1911.08947}, - year={2019} -} - -3. DTRB: -@inproceedings{baek2019wrong, - title={What is wrong with scene text recognition model comparisons? dataset and model analysis}, - author={Baek, Jeonghun and Kim, Geewook and Lee, Junyeop and Park, Sungrae and Han, Dongyoon and Yun, Sangdoo and Oh, Seong Joon and Lee, Hwalsuk}, - booktitle={Proceedings of the IEEE International Conference on Computer Vision}, - pages={4715--4723}, - year={2019} -} - -4. SAST: -@inproceedings{wang2019single, - title={A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning}, - author={Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming}, - booktitle={Proceedings of the 27th ACM International Conference on Multimedia}, - pages={1277--1285}, - year={2019} -} - -5. SRN: -@article{yu2020towards, - title={Towards Accurate Scene Text Recognition with Semantic Reasoning Networks}, - author={Yu, Deli and Li, Xuan and Zhang, Chengquan and Han, Junyu and Liu, Jingtuo and Ding, Errui}, - journal={arXiv preprint arXiv:2003.12294}, - year={2020} -} - -6. end2end-psl: -@inproceedings{sun2019chinese, - title={Chinese Street View Text: Large-scale Chinese Text Reading with Partially Supervised Learning}, - author={Sun, Yipeng and Liu, Jiaming and Liu, Wei and Han, Junyu and Ding, Errui and Liu, Jingtuo}, - booktitle={Proceedings of the IEEE International Conference on Computer Vision}, - pages={9086--9095}, - year={2019} -} -``` diff --git a/doc/config.md b/doc/config.md new file mode 100644 index 0000000000000000000000000000000000000000..282a6bfb15b546b52752c2e5ad763a09f9f428c8 --- /dev/null +++ b/doc/config.md @@ -0,0 +1,35 @@ +# 可选参数列表 + +以下列表可以通过`--help`查看 + +| FLAG | 支持脚本 | 用途 | 默认值 | 备注 | +| :----------------------: | :------------: | :---------------: | :--------------: | :-----------------: | +| -c | ALL | 指定配置文件 | None | **配置模块说明请参考 参数介绍** | +| -o | ALL | 设置配置文件里的参数内容 | None | 使用-o配置相较于-c选择的配置文件具有更高的优先级。例如：`-o Global.use_gpu=false` | + + +## 配置文件 Global 参数介绍 + +| 字段 | 用途 | 默认值 | 备注 | +| :----------------------: | :---------------------: | :--------------: | :--------------------: | +| algorithm | 设置算法 | CRNN | 选择模型，支持模型请参考[简介]() | +| use_gpu | 设置代码运行场所 | true | \ | +| epoch_num | 最大训练epoch数 | 3000 | \ | +| log_smooth_window | 滑动窗口大小 | 20 | \ | +| print_batch_step | 设置打印log间隔 | 10 | \ | +| save_model_dir | 设置模型保存路径 | output/rec_CRNN | \ | +| save_epoch_step | 设置模型保存间隔 | 3 | \ | +| eval_batch_step | 设置模型评估间隔 | 2000 | \ | +|train_batch_size_per_card | 设置训练时单卡batch size | 256 | \ | +| test_batch_size_per_card | 设置评估时单卡batch size | 256 | \ | +| image_shape | 设置输入图片尺寸 | [3, 32, 100] | \ | +| max_text_length | 设置文本最大长度 | 25 | \ | +| character_type | 设置字符类型 | ch | en/ch, en时将使用默认dict，ch时使用自定义dict| +| character_dict_path | 设置字典路径 | ./ppocr/utils/ic15_dict.txt | \ | +| loss_type | 设置 loss 类型 | ctc | 支持两种loss： ctc / attention | +| reader_yml | 设置reader配置文件 | ./configs/rec/rec_icdar15_reader.yml | \ | +| pretrain_weights | 加载预训练模型路径 | ./pretrain_models/CRNN/best_accuracy | \ | +| checkpoints | 加载模型参数路径 | None | 用于中断后重新训练 | +| save_inference_dir | inference model 保存路径 | None | 用于保存inference model | + + diff --git a/doc/detection.md b/doc/detection.md index 5e5501103777deb8de541d45c0ec02e6d108236f..1a0417ad5eddf009b92773bdc117e60fc343dbb8 100644 --- a/doc/detection.md +++ b/doc/detection.md @@ -8,8 +8,8 @@ icdar2015数据集可以从[官网](https://rrc.cvc.uab.es/?ch=4&com=downloads) 将下载到的数据集解压到工作目录下，假设解压在/PaddleOCR/train_data/ 下。另外，PaddleOCR将零散的标注文件整理成单独的标注文件 ，您可以通过wget的方式进行下载。 ``` -wget -P /PaddleOCR/train_data/ 训练标注文件链接 -wget -P /PaddleOCR/train_data/ 测试标注文件链接 +wget -P /PaddleOCR/train_data/ https://paddleocr.bj.bcebos.com/dataset%2Ftrain_icdar2015_label.txt +wget -P /PaddleOCR/train_data/ https://paddleocr.bj.bcebos.com/dataset%2Ftest_icdar2015_label.txt ``` 解压数据集和下载标注文件后，/PaddleOCR/train_data/ 有两个文件夹和两个文件，分别是： @@ -38,9 +38,9 @@ $transcription$表示当前文本框的文字，在文本检测任务中并不 ``` cd PaddleOCR/ # 下载MobileNetV3的预训练模型 -wget -P /PaddleOCR/pretrain_models/ 模型链接 +wget -P /PaddleOCR/pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x0_5_pretrained.tar # 下载ResNet50的预训练模型 -wget -P /PaddleOCR/pretrain_models/ 模型链接 +wget -P /PaddleOCR/pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_ssld_pretrained.tar ``` **启动训练** @@ -49,7 +49,7 @@ python3 tools/train.py -c configs/det/det_db_mv3.yml ``` 上述指令中，通过-c 选择训练使用configs/det/det_db_mv3.yml配置文件。 -有关配置文件的详细解释，请参考[链接]()。 +有关配置文件的详细解释，请参考[链接](./doc/config.md)。 您也可以通过-o参数在不需要修改yml文件的情况下，改变训练的参数，比如，调整训练的学习率为0.0001 ``` diff --git a/ppocr/utils/character.py b/ppocr/utils/character.py index 7dfeeef4c15b6bb95d37edc16b99f04ea5ffe020..b40750392291f271b26fef88e58844be9020d2ea 100755 --- a/ppocr/utils/character.py +++ b/ppocr/utils/character.py @@ -64,7 +64,7 @@ class CharacterOps(object): [sum(text_lengths)] = [text_index_0 + text_index_1 + ... + text_index_(n - 1)] length: length of each text. [batch_size] """ - if self.character_type == "en" or text.encode( 'UTF-8' ).isalpha(): + if self.character_type == "en": text = text.lower() text_list = [] diff --git a/ppocr/utils/ic15_dict.txt b/ppocr/utils/ic15_dict.txt index c1f9993dae8be61d968b602ddde5683b8e4ca959..71043689051fb5a2da516b2e005d1d9b0fdecfb3 100644 --- a/ppocr/utils/ic15_dict.txt +++ b/ppocr/utils/ic15_dict.txt @@ -1,3 +1,13 @@ +0 +1 +2 +3 +4 +5 +6 +7 +8 +9 a b c @@ -24,13 +34,3 @@ w x y z -0 -1 -2 -3 -4 -5 -6 -7 -8 -9