diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java index b474d8886a10746b8ac181085c62481dfe7a4229..b7f9138422fe9ae0fa63c9cebba49ecf117b5c2b 100644 --- a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java +++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java @@ -34,7 +34,7 @@ public class Predictor { protected String inputColorFormat = "BGR"; protected long[] inputShape = new long[]{1, 3, 960}; protected float[] inputMean = new float[]{0.485f, 0.456f, 0.406f}; - protected float[] inputStd = new float[]{1.0f / 0.229f, 1.0f / 0.224f, 1.0f / 0.225f}; + protected float[] inputStd = new float[]{0.229f, 0.224f, 0.225f}; protected float scoreThreshold = 0.1f; protected Bitmap inputImage = null; protected Bitmap outputImage = null; diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index f0c16618c0dd0b0f0bcc6a06d6b142a59d58e725..0efa63d002b435f07ad7c3aaf614e2471b4f6181 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -25,8 +25,8 @@ PaddleOCR开源的文本检测算法列表： 在ICDAR2015文本检测公开数据集上，算法效果如下： |模型|骨干网络|precision|recall|Hmean|下载链接| | --- | --- | --- | --- | --- | --- | -|EAST|ResNet50_vd|85.80%|86.71%|86.25%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)| -|EAST|MobileNetV3|79.42%|80.64%|80.03%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)| +|EAST|ResNet50_vd|88.71%|81.36%|84.88%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)| +|EAST|MobileNetV3|78.2%|79.1%|78.65%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)| |DB|ResNet50_vd|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)| |DB|MobileNetV3|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)| |SAST|ResNet50_vd|91.39%|83.77%|87.42%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)| diff --git a/doc/doc_ch/customize.md b/doc/doc_ch/customize.md index 5944bf08e4c4fac0da5a8e939936719e0385f4e1..653717391e699f208caf306bb7fa8ef21078434a 100644 --- a/doc/doc_ch/customize.md +++ b/doc/doc_ch/customize.md @@ -5,17 +5,21 @@ ## step1：训练文本检测模型 PaddleOCR提供了EAST、DB两种文本检测算法，均支持MobileNetV3、ResNet50_vd两种骨干网络，根据需要选择相应的配置文件，启动训练。例如，训练使用MobileNetV3作为骨干网络的DB检测模型（即超轻量模型使用的配置）： -``` + +```sh python3 tools/train.py -c configs/det/det_mv3_db.yml 2>&1 | tee det_db.log ``` + 更详细的数据准备和训练教程参考文档教程中[文本检测模型训练/评估/预测](./detection.md)。 ## step2：训练文本识别模型 PaddleOCR提供了CRNN、Rosetta、STAR-Net、RARE四种文本识别算法，均支持MobileNetV3、ResNet34_vd两种骨干网络，根据需要选择相应的配置文件，启动训练。例如，训练使用MobileNetV3作为骨干网络的CRNN识别模型（即超轻量模型使用的配置）： -``` + +```sh python3 tools/train.py -c configs/rec/rec_chinese_lite_train.yml 2>&1 | tee rec_ch_lite.log ``` + 更详细的数据准备和训练教程参考文档教程中[文本识别模型训练/评估/预测](./recognition.md)。 ## step3：模型串联预测 @@ -24,7 +28,8 @@ PaddleOCR提供了检测和识别模型的串联工具，可以将训练好的 在执行预测时，需要通过参数image_dir指定单张图像或者图像集合的路径、参数det_model_dir指定检测inference模型的路径和参数rec_model_dir指定识别inference模型的路径。可视化识别结果默认保存到 ./inference_results 文件夹里面。 -``` +```sh python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" ``` + 更多的文本检测、识别串联推理使用方式请参考文档教程中的[基于预测引擎推理](./inference.md)。 diff --git a/doc/doc_ch/knowledge_distillation.md b/doc/doc_ch/knowledge_distillation.md index 265703adc95507c33f00c44efd68560db4556c1c..0923fc9ccc03053bd0d2e6ba0606dc980145c917 100644 --- a/doc/doc_ch/knowledge_distillation.md +++ b/doc/doc_ch/knowledge_distillation.md @@ -1,8 +1,27 @@ + # 知识蒸馏 - ++ [知识蒸馏](#0) + + [1. 简介](#1) + - [1.1 知识蒸馏介绍](#11) + - [1.2 PaddleOCR知识蒸馏简介](#12) + + [2. 配置文件解析](#2) + + [2.1 识别配置文件解析](#21) + - [2.1.1 模型结构](#211) + - [2.1.2 损失函数](#212) + - [2.1.3 后处理](#213) + - [2.1.4 指标计算](#214) + - [2.1.5 蒸馏模型微调](#215) + + [2.2 检测配置文件解析](#22) + - [2.2.1 模型结构](#221) + - [2.2.2 损失函数](#222) + - [2.2.3 后处理](#223) + - [2.2.4 蒸馏指标计算](#224) + - [2.2.5 检测蒸馏模型Fine-tune](#225) + + ## 1. 简介 - + ### 1.1 知识蒸馏介绍 近年来，深度神经网络在计算机视觉、自然语言处理等领域被验证是一种极其有效的解决问题的方法。通过构建合适的神经网络，加以训练，最终网络模型的性能指标基本上都会超过传统算法。 @@ -13,6 +32,7 @@ 此外，在知识蒸馏任务中，也衍生出了互学习的模型训练方法，论文[Deep Mutual Learning](https://arxiv.org/abs/1706.00384)中指出，使用两个完全相同的模型在训练的过程中互相监督，可以达到比单个模型训练更好的效果。 + ### 1.2 PaddleOCR知识蒸馏简介 无论是大模型蒸馏小模型，还是小模型之间互相学习，更新参数，他们本质上是都是不同模型之间输出或者特征图(feature map)之间的相互监督，区别仅在于 (1) 模型是否需要固定参数。(2) 模型是否需要加载预训练模型。 @@ -30,17 +50,19 @@ PaddleOCR中集成了知识蒸馏的算法，具体地，有以下几个主要 通过知识蒸馏，在中英文通用文字识别任务中，不增加任何预测耗时的情况下，可以给模型带来3%以上的精度提升，结合学习率调整策略以及模型结构微调策略，最终提升提升超过5%。 - + ## 2. 配置文件解析 在知识蒸馏训练的过程中，数据预处理、优化器、学习率、全局的一些属性没有任何变化。模型结构、损失函数、后处理、指标计算等模块的配置文件需要进行微调。 下面以识别与检测的知识蒸馏配置文件为例，对知识蒸馏的训练与配置进行解析。 + ### 2.1 识别配置文件解析 配置文件在[ch_PP-OCRv2_rec_distillation.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml)。 + #### 2.1.1 模型结构 知识蒸馏任务中，模型结构配置如下所示。 @@ -176,6 +198,7 @@ Architecture: } ``` + #### 2.1.2 损失函数 知识蒸馏任务中，损失函数配置如下所示。 @@ -212,7 +235,7 @@ Loss: 关于`CombinedLoss`更加具体的实现可以参考: [combined_loss.py](../../ppocr/losses/combined_loss.py#L23)。关于`DistillationCTCLoss`等蒸馏损失函数更加具体的实现可以参考[distillation_loss.py](../../ppocr/losses/distillation_loss.py)。 - + #### 2.1.3 后处理 知识蒸馏任务中，后处理配置如下所示。 @@ -228,7 +251,7 @@ PostProcess: 关于`DistillationCTCLabelDecode`更加具体的实现可以参考: [rec_postprocess.py](../../ppocr/postprocess/rec_postprocess.py#L128) - + #### 2.1.4 指标计算 知识蒸馏任务中，指标计算配置如下所示。 @@ -245,7 +268,7 @@ Metric: 关于`DistillationMetric`更加具体的实现可以参考: [distillation_metric.py](../../ppocr/metrics/distillation_metric.py#L24)。 - + #### 2.1.5 蒸馏模型微调 对蒸馏得到的识别蒸馏进行微调有2种方式。 @@ -279,7 +302,7 @@ paddle.save(s_params, "ch_PP-OCRv2_rec_train/student.pdparams") 转化完成之后，使用[ch_PP-OCRv2_rec.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml)，修改预训练模型的路径（为导出的`student.pdparams`模型路径）以及自己的数据路径，即可进行模型微调。 - + ### 2.2 检测配置文件解析 检测模型蒸馏的配置文件在PaddleOCR/configs/det/ch_PP-OCRv2/目录下，包含三个蒸馏配置文件： @@ -287,7 +310,7 @@ paddle.save(s_params, "ch_PP-OCRv2_rec_train/student.pdparams") - ch_PP-OCRv2_det_dml.yml，采用DML的蒸馏，两个Student模型互蒸馏的方法 - ch_PP-OCRv2_det_distill.yml，采用Teacher大模型蒸馏小模型Student的方法 - + #### 2.2.1 模型结构 知识蒸馏任务中，模型结构配置如下所示： @@ -419,7 +442,8 @@ Architecture: } ``` -#### 2.1.2 损失函数 + +#### 2.2.2 损失函数 知识蒸馏任务中，检测ch_PP-OCRv2_det_distill.yml蒸馏损失函数配置如下所示。 @@ -484,8 +508,8 @@ Loss: 关于`DistillationDilaDBLoss`更加具体的实现可以参考: [distillation_loss.py](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.4/ppocr/losses/distillation_loss.py#L185)。关于`DistillationDBLoss`等蒸馏损失函数更加具体的实现可以参考[distillation_loss.py](https://github.com/PaddlePaddle/PaddleOCR/blob/04c44974b13163450dfb6bd2c327863f8a194b3c/ppocr/losses/distillation_loss.py?_pjax=%23js-repo-pjax-container%2C%20div%5Bitemtype%3D%22http%3A%2F%2Fschema.org%2FSoftwareSourceCode%22%5D%20main%2C%20%5Bdata-pjax-container%5D#L148)。 - -#### 2.1.3 后处理 + +#### 2.2.3 后处理 知识蒸馏任务中，检测蒸馏后处理配置如下所示。 @@ -504,7 +528,8 @@ PostProcess: 关于`DistillationDBPostProcess`更加具体的实现可以参考: [db_postprocess.py](../../ppocr/postprocess/db_postprocess.py#L195) -#### 2.1.4 蒸馏指标计算 + +#### 2.2.4 蒸馏指标计算 知识蒸馏任务中，检测蒸馏指标计算配置如下所示。 @@ -518,8 +543,8 @@ Metric: 由于蒸馏需要包含多个网络，甚至多个Student网络，在计算指标的时候只需要计算一个Student网络的指标即可，`key`字段设置为`Student`则表示只计算`Student`网络的精度。 - -#### 2.1.5 检测蒸馏模型finetune + +#### 2.2.5 检测蒸馏模型finetune 检测蒸馏有三种方式： - 采用ch_PP-OCRv2_det_distill.yml，Teacher模型设置为PaddleOCR提供的模型或者您训练好的大模型 diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md index 4d521c9cd35d4dbde9b8c83d8a7e5ebbd84e59c9..8d27613ba88660d760cfe1dcd3a1872ece1a51ad 100755 --- a/doc/doc_en/algorithm_overview_en.md +++ b/doc/doc_en/algorithm_overview_en.md @@ -30,8 +30,8 @@ On the ICDAR2015 dataset, the text detection result is as follows: |Model|Backbone|Precision|Recall|Hmean|Download link| | --- | --- | --- | --- | --- | --- | -|EAST|ResNet50_vd|85.80%|86.71%|86.25%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)| -|EAST|MobileNetV3|79.42%|80.64%|80.03%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)| +|EAST|ResNet50_vd|88.71%|81.36%|84.88%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)| +|EAST|MobileNetV3|78.2%|79.1%|78.65%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)| |DB|ResNet50_vd|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)| |DB|MobileNetV3|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)| |SAST|ResNet50_vd|91.39%|83.77%|87.42%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)| diff --git a/doc/doc_en/knowledge_distillation_en.md b/doc/doc_en/knowledge_distillation_en.md new file mode 100755 index 0000000000000000000000000000000000000000..1db9faef5c97cacbba36fdb42807924e9c7a53cf --- /dev/null +++ b/doc/doc_en/knowledge_distillation_en.md @@ -0,0 +1,592 @@ + +# Knowledge Distillation + ++ [Knowledge Distillation](#0) + + [1. Introduction](#1) + - [1.1 Introduction to Knowledge Distillation](#11) + - [1.2 Introduction to PaddleOCR Knowledge Distillation](#12) + + [2. Configuration File Analysis](#2) + + [2.1 Recognition Model Configuration File Analysis](#21) + - [2.1.1 Model Structure](#211) + - [2.1.2 Loss Function ](#212) + - [2.1.3 Post-processing](#213) + - [2.1.4 Metric Calculation](#214) + - [2.1.5 Fine-tuning Distillation Model](#215) + + [2.2 Detection Model Configuration File Analysis](#22) + - [2.2.1 Model Structure](#221) + - [2.2.2 Loss Function](#222) + - [2.2.3 Post-processing](#223) + - [2.2.4 Metric Calculation](#224) + - [2.2.5 Fine-tuning Distillation Model](#225) + + + + +## 1. Introduction + +### 1.1 Introduction to Knowledge Distillation + +In recent years, deep neural networks have been proved to be an extremely effective method for solving problems in the fields of computer vision and natural language processing. +By constructing a suitable neural network and training it, the performance metrics of the final network model will basically exceed the traditional algorithm. +When the amount of data is large enough, increasing the amount of parameters by constructing a reasonable network model can significantly improve the performance of the model, +but this brings about the problem of a sharp increase in the complexity of the model. Large models are more expensive to use in actual scenarios. +Deep neural networks generally have more parameter redundancy. At present, there are several main methods to compress the model and reduce the amount of its parameters. +Such as pruning, quantification, knowledge distillation, etc., where knowledge distillation refers to the use of teacher models to guide student models to learn specific tasks, +to ensure that the small model obtains a relatively large performance improvement under the condition of unchanged parameters. +In addition, in the knowledge distillation task, a mutual learning model training method was also derived. +The paper [Deep Mutual Learning](https://arxiv.org/abs/1706.00384) pointed out that using two identical models to supervise each other during the training process can achieve better results than a single model training. + + +### 1.2 Introduction to PaddleOCR Knowledge Distillation + +Whether it is a large model distilling a small model, or a small model learning from each other and updating parameters, +they are essentially the output between different models or mutual supervision between feature maps. +The only difference is (1) whether the model requires fixed parameters. (2) Whether the model needs to be loaded with a pre-trained model. +For the case where a large model distills a small model, the large model generally needs to load the pre-trained model and fix the parameters. +For the situation where small models distill each other, the small models generally do not load the pre-trained model, and the parameters are also in a learnable state. + +In the task of knowledge distillation, it is not only the distillation between two models, but also the situation where multiple models learn from each other. +Therefore, in the knowledge distillation code framework, it is also necessary to support this type of distillation method. + +The algorithm of knowledge distillation is integrated in PaddleOCR. Specifically, it has the following main features: +- It supports mutual learning of any network, and does not require the sub-network structure to be completely consistent or to have a pre-trained model. At the same time, there is no limit to the number of sub-networks, just add it in the configuration file. +- Support arbitrarily configuring the loss function through the configuration file, not only can use a certain loss, but also a combination of multiple losses. +- Support all model-related environments such as knowledge distillation training, prediction, evaluation, and export, which is convenient for use and deployment. + +Through knowledge distillation, in the common Chinese and English text recognition task, without adding any time-consuming prediction, +the accuracy of the model can be improved by more than 3%. Combining the learning rate adjustment strategy and the model structure fine-tuning strategy, +the final improvement is more than 5%. + + +## 2. Configuration File Analysis + +In the process of knowledge distillation training, there is no change in data preprocessing, optimizer, learning rate, and some global attributes. +The configuration files of the model structure, loss function, post-processing, metric calculation and other modules need to be fine-tuned. + +The following takes the knowledge distillation configuration file for recognition and detection as an example to analyze the training and configuration of knowledge distillation. + + +### 2.1 Recognition Model Configuration File Analysis + +The configuration file is in [ch_PP-OCRv2_rec_distillation.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml). + + +#### 2.1.1 Model Structure + +In the knowledge distillation task, the model structure configuration is as follows. +```yaml +Architecture: + model_type: &model_type "rec" # Model category, recognition, detection, etc. + name: DistillationModel # Structure name, in the distillation task, it is DistillationModel + algorithm: Distillation # Algorithm name + Models: # Model, including the configuration information of the subnet + Teacher: # The name of the subnet, it must include at least the `pretrained` and `freeze_params` parameters, and the other parameters are the construction parameters of the subnet + pretrained: # Does this sub-network need to load pre-training weights + freeze_params: false # Do you need fixed parameters + return_all_feats: true # Do you need to return all features, if it is False, only the final output is returned + model_type: *model_type # Model category + algorithm: CRNN # The algorithm name of the sub-network. The remaining parameters of the sub-network are consistent with the general model training configuration + Transform: + Backbone: + name: MobileNetV1Enhance + scale: 0.5 + Neck: + name: SequenceEncoder + encoder_type: rnn + hidden_size: 64 + Head: + name: CTCHead + mid_channels: 96 + fc_decay: 0.00002 + Student: # Another sub-network, here is a distillation example of DML, the two sub-networks have the same structure, and both need to learn parameters + pretrained: # The following parameters are the same as above + freeze_params: false + return_all_feats: true + model_type: *model_type + algorithm: CRNN + Transform: + Backbone: + name: MobileNetV1Enhance + scale: 0.5 + Neck: + name: SequenceEncoder + encoder_type: rnn + hidden_size: 64 + Head: + name: CTCHead + mid_channels: 96 + fc_decay: 0.00002 +``` + +If you want to add more sub-networks for training, you can also add the corresponding fields in the configuration file according to the way of adding `Student` and `Teacher`. +For example, if you want 3 models to supervise each other and train together, then `Architecture` can be written in the following format. + +```yaml +Architecture: + model_type: &model_type "rec" + name: DistillationModel + algorithm: Distillation + Models: + Teacher: + pretrained: + freeze_params: false + return_all_feats: true + model_type: *model_type + algorithm: CRNN + Transform: + Backbone: + name: MobileNetV1Enhance + scale: 0.5 + Neck: + name: SequenceEncoder + encoder_type: rnn + hidden_size: 64 + Head: + name: CTCHead + mid_channels: 96 + fc_decay: 0.00002 + Student: + pretrained: + freeze_params: false + return_all_feats: true + model_type: *model_type + algorithm: CRNN + Transform: + Backbone: + name: MobileNetV1Enhance + scale: 0.5 + Neck: + name: SequenceEncoder + encoder_type: rnn + hidden_size: 64 + Head: + name: CTCHead + mid_channels: 96 + fc_decay: 0.00002 + Student2: # The new sub-network introduced in the knowledge distillation task, the configuration is the same as above + pretrained: + freeze_params: false + return_all_feats: true + model_type: *model_type + algorithm: CRNN + Transform: + Backbone: + name: MobileNetV1Enhance + scale: 0.5 + Neck: + name: SequenceEncoder + encoder_type: rnn + hidden_size: 64 + Head: + name: CTCHead + mid_channels: 96 + fc_decay: 0.00002 +``` + +When the model is finally trained, it contains 3 sub-networks: `Teacher`, `Student`, `Student2`. + +The specific implementation code of the `DistillationModel` class can refer to [distillation_model.py](../../ppocr/modeling/architectures/distillation_model.py). +The final model output is a dictionary, the key is the name of all the sub-networks, for example, here are `Student` and `Teacher`, and the value is the output of the corresponding sub-network, +which can be `Tensor` (only the last layer of the network is returned) and `dict` (also returns the characteristic information in the middle). +In the recognition task, in order to add more loss functions and ensure the scalability of the distillation method, the output of each sub-network is saved as a `dict`, which contains the sub-module output. +Take the recognition model as an example. The output result of each sub-network is `dict`, the key contains `backbone_out`, `neck_out`, `head_out`, and `value` is the tensor of the corresponding module. Finally, for the above configuration file, `DistillationModel` The output format is as follows. + +```json +{ + "Teacher": { + "backbone_out": tensor, + "neck_out": tensor, + "head_out": tensor, + }, + "Student": { + "backbone_out": tensor, + "neck_out": tensor, + "head_out": tensor, + } +} +``` + + +#### 2.1.2 Loss Function + +In the knowledge distillation task, the loss function configuration is as follows. + +```yaml +Loss: + name: CombinedLoss # Loss function name + loss_config_list: # List of loss function configuration files, mandatory functions for CombinedLoss + - DistillationCTCLoss: # CTC loss function based on distillation, inherited from standard CTC loss + weight: 1.0 # The weight of the loss function. In loss_config_list, each loss function must include this field + model_name_list: ["Student", "Teacher"] # For the prediction results of the distillation model, extract the output of these two sub-networks and calculate the CTC loss with gt + key: head_out # In the sub-network output dict, take the corresponding tensor + - DistillationDMLLoss: # DML loss function, inherited from the standard DMLLoss + weight: 1.0 + act: "softmax" # Activation function, use it to process the input, can be softmax, sigmoid or None, the default is None + model_name_pairs: # The subnet name pair used to calculate DML loss. If you want to calculate the DML loss of other subnets, you can continue to add it below the list + - ["Student", "Teacher"] + key: head_out + - DistillationDistanceLoss: # Distilled distance loss function + weight: 1.0 + mode: "l2" # Support l1, l2 or smooth_l1 + model_name_pairs: # Calculate the distance loss of the subnet name pair + - ["Student", "Teacher"] + key: backbone_out +``` + +Among the above loss functions, all distillation loss functions are inherited from the standard loss function class. +The main functions are: Analyze the output of the distillation model, find the intermediate node (tensor) used to calculate the loss, +and then use the standard loss function class to calculate. + +Taking the above configuration as an example, the final distillation training loss function contains the following three parts. + +- The final output `head_out` of `Student` and `Teacher` calculates the CTC loss with gt (loss weight equals 1.0). Here, because both sub-networks need to update the parameters, both of them need to calculate the loss with gt. +- DML loss between `Student` and `Teacher`'s final output `head_out` (loss weight equals 1.0). +- L2 loss between `Student` and `Teacher`'s backbone network output `backbone_out` (loss weight equals 1.0). + +For more specific implementation of `CombinedLoss`, please refer to: [combined_loss.py](../../ppocr/losses/combined_loss.py#L23). +For more specific implementations of distillation loss functions such as `DistillationCTCLoss`, please refer to [distillation_loss.py](../../ppocr/losses/distillation_loss.py) + + + +#### 2.1.3 Post-processing + +In the knowledge distillation task, the post-processing configuration is as follows. + +```yaml +PostProcess: + name: DistillationCTCLabelDecode # CTC decoding post-processing of distillation tasks, inherited from the standard CTCLabelDecode class + model_name: ["Student", "Teacher"] # For the prediction results of the distillation model, extract the outputs of these two sub-networks and decode them + key: head_out # Take the corresponding tensor in the subnet output dict +``` + +Taking the above configuration as an example, the CTC decoding output of the two sub-networks `Student` and `Teahcer` will be calculated at the same time. +Among them, `key` is the name of the subnet, and `value` is the list of subnets. + +For more specific implementation of `DistillationCTCLabelDecode`, please refer to: [rec_postprocess.py](../../ppocr/postprocess/rec_postprocess.py#L128) + + + +#### 2.1.4 Metric Calculation + +In the knowledge distillation task, the metric calculation configuration is as follows. + +```yaml +Metric: + name: DistillationMetric # CTC decoding post-processing of distillation tasks, inherited from the standard CTCLabelDecode class + base_metric_name: RecMetric # The base class of indicator calculation. For the output of the model, the indicator will be calculated based on this class + main_indicator: acc # The name of the indicator + key: "Student" # Select the main_indicator of this subnet as the criterion for saving the best model +``` + +Taking the above configuration as an example, the accuracy metric of the `Student` subnet will be used as the judgment metric for saving the best model. +At the same time, the accuracy metric of all subnets will be printed out in the log. + +For more specific implementation of `DistillationMetric`, please refer to: [distillation_metric.py](../../ppocr/metrics/distillation_metric.py#L24). + + + +#### 2.1.5 Fine-tuning Distillation Model + +There are two ways to fine-tune the recognition distillation task. + +1. Fine-tuning based on knowledge distillation: this situation is relatively simple, download the pre-trained model. Then configure the pre-training model path and your own data path in [ch_PP-OCRv2_rec_distillation.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml) to perform fine-tuning training of the model. +2. Do not use knowledge distillation in fine-tuning: In this case, you need to first extract the student model parameters from the pre-training model. The specific steps are as follows. + +- First download the pre-trained model and unzip it. +```shell +wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar +tar -xf ch_PP-OCRv2_rec_train.tar +``` + +- Then use python to extract the student model parameters + +```python +import paddle +# Load the pre-trained model +all_params = paddle.load("ch_PP-OCRv2_rec_train/best_accuracy.pdparams") +# View the keys of the weight parameter +print(all_params.keys()) +# Weight extraction of student model +s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key} +# View the keys of the weight parameters of the student model +print(s_params.keys()) +# Save weight parameters +paddle.save(s_params, "ch_PP-OCRv2_rec_train/student.pdparams") +``` + +After the extraction is complete, use [ch_PP-OCRv2_rec.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml) to modify the path of the pre-trained model (the path of the exported `student.pdparams` model) and your own data path to fine-tune the model. + + +### 2.2 Detection Model Configuration File Analysis + +The configuration file of the detection model distillation is in the ```PaddleOCR/configs/det/ch_PP-OCRv2/``` directory, which contains three distillation configuration files: + +- ```ch_PP-OCRv2_det_cml.yml```, Use one large model to distill two small models, and the two small models learn from each other +- ```ch_PP-OCRv2_det_dml.yml```, Method of mutual distillation of two student models +- ```ch_PP-OCRv2_det_distill.yml```, The method of using large teacher model to distill small student model + + +#### 2.2.1 Model Structure + +In the knowledge distillation task, the model structure configuration is as follows: +``` +Architecture: + name: DistillationModel # Structure name, in the distillation task, it is DistillationModel + algorithm: Distillation # Algorithm name + Models: # Model, including the configuration information of the subnet + Student: # The name of the subnet, it must include at least the `pretrained` and `freeze_params` parameters, and the other parameters are the construction parameters of the subnet + pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained # Does this sub-network need to load pre-training weights + freeze_params: false # Do you need fixed parameters + return_all_feats: false # Do you need to return all features, if it is False, only the final output is returned + model_type: det + algorithm: DB + Backbone: + name: MobileNetV3 + scale: 0.5 + model_name: large + disable_se: True + Neck: + name: DBFPN + out_channels: 96 + Head: + name: DBHead + k: 50 + Teacher: # Another sub-network, here is a distillation example of a large model distill a small model + pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy + freeze_params: true # The Teacher model is well-trained and does not need to participate in training + return_all_feats: false + model_type: det + algorithm: DB + Transform: + Backbone: + name: ResNet + layers: 18 + Neck: + name: DBFPN + out_channels: 256 + Head: + name: DBHead + k: 50 + +``` +If DML is used, that is, the method of two small models learning from each other, the Teacher network structure in the above configuration file needs to be set to the same configuration as the Student model. +Refer to the configuration file for details. [ch_PP-OCRv2_det_dml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.4/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml) + + +The following describes the configuration file parameters [ch_PP-OCRv2_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.4/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml): + +``` +Architecture: + name: DistillationModel + algorithm: Distillation + model_type: det + Models: + Teacher: # Teacher model configuration of CML distillation + pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy + freeze_params: true # Teacher does not train + return_all_feats: false + model_type: det + algorithm: DB + Transform: + Backbone: + name: ResNet + layers: 18 + Neck: + name: DBFPN + out_channels: 256 + Head: + name: DBHead + k: 50 + Student: # Student model configuration for CML distillation + pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained + freeze_params: false + return_all_feats: false + model_type: det + algorithm: DB + Backbone: + name: MobileNetV3 + scale: 0.5 + model_name: large + disable_se: True + Neck: + name: DBFPN + out_channels: 96 + Head: + name: DBHead + k: 50 + Student2: # Student2 model configuration for CML distillation + pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained + freeze_params: false + return_all_feats: false + model_type: det + algorithm: DB + Transform: + Backbone: + name: MobileNetV3 + scale: 0.5 + model_name: large + disable_se: True + Neck: + name: DBFPN + out_channels: 96 + Head: + name: DBHead + k: 50 + +``` + +The specific implementation code of the distillation model `DistillationModel` class can refer to [distillation_model.py](../../ppocr/modeling/architectures/distillation_model.py). + +The final model output is a dictionary, the key is the name of all the sub-networks, for example, here are `Student` and `Teacher`, and the value is the output of the corresponding sub-network, +which can be `Tensor` (only the last layer of the network is returned) and `dict` (also returns the characteristic information in the middle). + +In the distillation task, in order to facilitate the addition of the distillation loss function, the output of each network is saved as a `dict`, which contains the sub-module output. +The key contains `backbone_out`, `neck_out`, `head_out`, and `value` is the tensor of the corresponding module. Finally, for the above configuration file, the output format of `DistillationModel` is as follows. + +```json +{ + "Teacher": { + "backbone_out": tensor, + "neck_out": tensor, + "head_out": tensor, + }, + "Student": { + "backbone_out": tensor, + "neck_out": tensor, + "head_out": tensor, + } +} +``` + + +#### 2.2.2 Loss Function + +In the task of detection knowledge distillation ```ch_PP-OCRv2_det_distill.yml````, the distillation loss function configuration is as follows. +```yaml +Loss: + name: CombinedLoss # Loss function name + loss_config_list: # List of loss function configuration files, mandatory functions for CombinedLoss + - DistillationDilaDBLoss: # DB loss function based on distillation, inherited from standard DBloss + weight: 1.0 # The weight of the loss function. In loss_config_list, each loss function must include this field + model_name_pairs: # Extract the output of these two sub-networks and calculate the loss between them + - ["Student", "Teacher"] + key: maps # In the sub-network output dict, take the corresponding tensor + balance_loss: true # The following parameters are the configuration parameters of standard DBloss + main_loss_type: DiceLoss + alpha: 5 + beta: 10 + ohem_ratio: 3 + - DistillationDBLoss: # Used to calculate the loss between Student and GT + weight: 1.0 + model_name_list: ["Student"] # The model name only has Student, which means that the loss between Student and GT is calculated + name: DBLoss + balance_loss: true + main_loss_type: DiceLoss + alpha: 5 + beta: 10 + ohem_ratio: 3 +``` + +Similarly, distillation loss function configuration(`ch_PP-OCRv2_det_cml.yml`) is shown below. Compared with the loss function configuration of ch_PP-OCRv2_det_distill.yml, there are three changes: +```yaml +Loss: + name: CombinedLoss + loss_config_list: + - DistillationDilaDBLoss: + weight: 1.0 + model_name_pairs: + - ["Student", "Teacher"] + - ["Student2", "Teacher"] # 1. Calculate the loss of two Student and Teacher + key: maps + balance_loss: true + main_loss_type: DiceLoss + alpha: 5 + beta: 10 + ohem_ratio: 3 + - DistillationDMLLoss: # 2. Add to calculate the loss between two students + model_name_pairs: + - ["Student", "Student2"] + maps_name: "thrink_maps" + weight: 1.0 + # act: None + key: maps + - DistillationDBLoss: + weight: 1.0 + model_name_list: ["Student", "Student2"] # 3. Calculate the loss between two students and GT + balance_loss: true + main_loss_type: DiceLoss + alpha: 5 + beta: 10 + ohem_ratio: 3 +``` + +For more specific implementation of `DistillationDilaDBLoss`, please refer to: [distillation_loss.py](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.4/ppocr/losses/distillation_loss.py#L185). +For more specific implementations of distillation loss functions such as `DistillationDBLoss`, please refer to: [distillation_loss.py](https://github.com/PaddlePaddle/PaddleOCR/blob/04c44974b13163450dfb6bd2c327863f8a194b3c/ppocr/losses/distillation_loss.py?_pjax=%23js-repo-pjax-container%2C%20div%5Bitemtype%3D%22http%3A%2F%2Fschema.org%2FSoftwareSourceCode%22%5D%20main%2C%20%5Bdata-pjax-container%5D#L148) + + +#### 2.2.3 Post-processing + +In the task of detecting knowledge distillation, the post-processing configuration of detecting distillation is as follows. + +```yaml +PostProcess: + name: DistillationDBPostProcess # The CTC decoding post-processing of the DB detection distillation task, inherited from the standard DBPostProcess class + model_name: ["Student", "Student2", "Teacher"] # Extract the output of multiple sub-networks and decode them. The network that does not require post-processing is not set in model_name + thresh: 0.3 + box_thresh: 0.6 + max_candidates: 1000 + unclip_ratio: 1.5 +``` + +Taking the above configuration as an example, the output of the three subnets `Student`, `Student2` and `Teacher` will be calculated at the same time for post-processing calculations. +Since there are multiple inputs, there are also multiple outputs returned by post-processing. +For a more specific implementation of `DistillationDBPostProcess`, please refer to: [db_postprocess.py](../../ppocr/postprocess/db_postprocess.py#L195) + + +#### 2.2.4 Metric Calculation +In the knowledge distillation task, the metric calculation configuration is as follows. +```yaml +Metric: + name: DistillationMetric + base_metric_name: DetMetric + main_indicator: hmean + key: "Student" +``` + +Since distillation needs to include multiple networks, only one network metrics needs to be calculated when calculating the metrics. +The `key` field is set to `Student`, it means that only the metrics of the `Student` network is calculated. +Model Structure + + +#### 2.2.5 Fine-tuning Distillation Model + +There are three ways to fine-tune the detection distillation task: +- `ch_PP-OCRv2_det_distill.yml`, The teacher model is set to the model provided by PaddleOCR or the large model you have trained. +- `ch_PP-OCRv2_det_cml.yml`, Use cml distillation. Similarly, the Teacher model is set to the model provided by PaddleOCR or the large model you have trained. +- `ch_PP-OCRv2_det_dml.yml`, Distillation using DML. The method of mutual distillation of the two Student models has an accuracy improvement of about 1.7% on the data set used by PaddleOCR. + +In fine-tune, you need to set the pre-trained model to be loaded in the `pretrained` parameter of the network structure. + +In terms of accuracy improvement, `cml` > `dml` > `distill`. When the amount of data is insufficient or the accuracy of the teacher model is similar to that of the student, this conclusion may change. + +In addition, since the distillation pre-training model provided by PaddleOCR contains multiple model parameters, if you want to extract the parameters of the student model, you can refer to the following code: +```sh +# Download the parameters of the distillation training model +wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar +``` + +```python +import paddle +# Load the pre-trained model +all_params = paddle.load("ch_PP-OCRv2_det_distill_train/best_accuracy.pdparams") +# View the keys of the weight parameter +print(all_params.keys()) +# Extract the weights of the student model +s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key} +# View the keys of the weight parameters of the student model +print(s_params.keys()) +# Save +paddle.save(s_params, "ch_PP-OCRv2_det_distill_train/student.pdparams") +``` + +Finally, the parameters of the student model will be saved in `ch_PP-OCRv2_det_distill_train/student.pdparams` for the fine-tune of the model. diff --git a/paddleocr.py b/paddleocr.py index 733c83d1b4faa23212e7186148a5a9e1154ba891..a530d7868523019920652ec4611832e00fc4da3a 100644 --- a/paddleocr.py +++ b/paddleocr.py @@ -42,7 +42,7 @@ __all__ = [ ] SUPPORT_DET_MODEL = ['DB'] -VERSION = '2.3.0.2' +VERSION = '2.4' SUPPORT_REC_MODEL = ['CRNN'] BASE_DIR = os.path.expanduser("~/.paddleocr/") diff --git a/ppstructure/README.md b/ppstructure/README.md index a09a43299b11dccf99897d5a6c69704191253aaf..abcc5281aa24c56e6a3f22e377de816b0bd66712 100644 --- a/ppstructure/README.md +++ b/ppstructure/README.md @@ -1,187 +1,140 @@ English | [简体中文](README_ch.md) -# PP-Structure +- [1. Introduction](#1) +- [2. Update log](#2) +- [3. Features](#3) +- [4. Results](#4) + * [4.1 Layout analysis and table recognition](#41) + * [4.2 DOC-VQA](#42) +- [5. Quick start](#5) +- [6. PP-Structure System](#6) + * [6.1 Layout analysis and table recognition](#61) + * [6.2 DOC-VQA](#62) +- [7. Model List](#7) -PP-Structure is an OCR toolkit that can be used for complex documents analysis. The main features are as follows: -- Support the layout analysis of documents, divide the documents into 5 types of areas **text, title, table, image and list** (conjunction with Layout-Parser) -- Support to extract the texts from the text, title, picture and list areas (used in conjunction with PP-OCR) -- Support to extract excel files from the table areas -- Support python whl package and command line usage, easy to use -- Support custom training for layout analysis and table structure tasks + -## 1. Visualization - - +## 1. Introduction +PP-Structure is an OCR toolkit that can be used for document analysis and processing with complex structures, designed to help developers better complete document understanding tasks + -## 2. Installation +## 2. Update log +* 2021.12.07 add [DOC-VQA SER and RE tasks](vqa/README.md)。 -### 2.1 Install requirements + -- **（1) Install PaddlePaddle** +## 3. Features -```bash -pip3 install --upgrade pip +The main features of PP-Structure are as follows: -# GPU -python3 -m pip install paddlepaddle-gpu==2.1.1 -i https://mirror.baidu.com/pypi/simple - -# CPU - python3 -m pip install paddlepaddle==2.1.1 -i https://mirror.baidu.com/pypi/simple +- Support the layout analysis of documents, divide the documents into 5 types of areas **text, title, table, image and list** (conjunction with Layout-Parser) +- Support to extract the texts from the text, title, picture and list areas (used in conjunction with PP-OCR) +- Support to extract excel files from the table areas +- Support python whl package and command line usage, easy to use +- Support custom training for layout analysis and table structure tasks +- Support Document Visual Question Answering (DOC-VQA) tasks: Semantic Entity Recognition (SER) and Relation Extraction (RE) -``` -For more，refer [Installation](https://www.paddlepaddle.org.cn/install/quick) . -- **(2) Install Layout-Parser** + -```bash -pip3 install -U https://paddleocr.bj.bcebos.com/whl/layoutparser-0.0.0-py3-none-any.whl -``` +## 4. Results -### 2.2 Install PaddleOCR（including PP-OCR and PP-Structure） + -- **（1) PIP install PaddleOCR whl package（inference only）** +### 4.1 Layout analysis and table recognition -```bash -pip install "paddleocr>=2.2" -``` + -- **（2) Clone PaddleOCR（Inference+training）** +The figure shows the pipeline of layout analysis + table recognition. The image is first divided into four areas of image, text, title and table by layout analysis, and then OCR detection and recognition is performed on the three areas of image, text and title, and the table is performed table recognition, where the image will also be stored for use. -```bash -git clone https://github.com/PaddlePaddle/PaddleOCR -``` + +### 4.2 DOC-VQA -## 3. Quick Start +* SER -### 3.1 Use by command line +![](./vqa/images/result_ser/zh_val_0_ser.jpg) | ![](./vqa/images/result_ser/zh_val_42_ser.jpg) +---|--- -```bash -paddleocr --image_dir=../doc/table/1.png --type=structure -``` +Different colored boxes in the figure represent different categories. For xfun dataset, there are three categories: query, answer and header: -### 3.2 Use by python API +* Dark purple: header +* Light purple: query +* Army green: answer -```python -import os -import cv2 -from paddleocr import PPStructure,draw_structure_result,save_structure_res +The corresponding category and OCR recognition results are also marked at the top left of the OCR detection box. -table_engine = PPStructure(show_log=True) -save_folder = './output/table' -img_path = '../doc/table/1.png' -img = cv2.imread(img_path) -result = table_engine(img) -save_structure_res(result, save_folder,os.path.basename(img_path).split('.')[0]) +* RE -for line in result: - line.pop('img') - print(line) +![](./vqa/images/result_re/zh_val_21_re.jpg) | ![](./vqa/images/result_re/zh_val_40_re.jpg) +---|--- -from PIL import Image -font_path = '../doc/fonts/simfang.ttf' -image = Image.open(img_path).convert('RGB') -im_show = draw_structure_result(image, result,font_path=font_path) -im_show = Image.fromarray(im_show) -im_show.save('result.jpg') -``` -### 3.3 Returned results format -The returned results of PP-Structure is a list composed of a dict, an example is as follows +In the figure, the red box represents the question, the blue box represents the answer, and the question and answer are connected by green lines. The corresponding category and OCR recognition results are also marked at the top left of the OCR detection box. -```shell -[ - { 'type': 'Text', - 'bbox': [34, 432, 345, 462], - 'res': ([[36.0, 437.0, 341.0, 437.0, 341.0, 446.0, 36.0, 447.0], [41.0, 454.0, 125.0, 453.0, 125.0, 459.0, 41.0, 460.0]], - [('Tigure-6. The performance of CNN and IPT models using difforen', 0.90060663), ('Tent ', 0.465441)]) - } -] -``` -The description of each field in dict is as follows -| Parameter | Description | -| --------------- | -------------| -|type|Type of image area| -|bbox|The coordinates of the image area in the original image, respectively [left upper x, left upper y, right bottom x, right bottom y]| -|res|OCR or table recognition result of image area。
Table: HTML string of the table;
OCR: A tuple containing the detection coordinates and recognition results of each single line of text| + +## 5. Quick start -### 3.4 Parameter description： +Start from [Quick Installation](./docs/quickstart.md) -| Parameter | Description | Default value | -| --------------- | ---------------------------------------- | ------------------------------------------- | -| output | The path where excel and recognition results are saved | ./output/table | -| table_max_len | The long side of the image is resized in table structure model | 488 | -| table_model_dir | inference model path of table structure model | None | -| table_char_type | dict path of table structure model | ../ppocr/utils/dict/table_structure_dict.tx | + -Most of the parameters are consistent with the paddleocr whl package, see [doc of whl](../doc/doc_en/whl_en.md) +## 6. PP-Structure System -After running, each image will have a directory with the same name under the directory specified in the output field. Each table in the picture will be stored as an excel and figure area will be cropped and saved, the excel and image file name will be the coordinates of the table in the image. + -## 4. PP-Structure Pipeline -![pipeline](../doc/table/pipeline_en.jpg) +### 6.1 Layout analysis and table recognition -In PP-Structure, the image will be analyzed by layoutparser first. In the layout analysis, the area in the image will be classified, including **text, title, image, list and table** 5 categories. For the first 4 types of areas, directly use the PP-OCR to complete the text detection and recognition. The table area will be converted to an excel file of the same table style via Table OCR. +![pipeline](../doc/table/pipeline.jpg) -### 4.1 LayoutParser +In PP-Structure, the image will be divided into 5 types of areas **text, title, image list and table**. For the first 4 types of areas, directly use PP-OCR system to complete the text detection and recognition. For the table area, after the table structuring process, the table in image is converted into an Excel file with the same table style. -Layout analysis divides the document data into regions, including the use of Python scripts for layout analysis tools, extraction of special category detection boxes, performance indicators, and custom training layout analysis models. For details, please refer to [document](layout/README_en.md). +#### 6.1.1 Layout analysis -### 4.2 Table Recognition +Layout analysis classifies image by region, including the use of Python scripts of layout analysis tools, extraction of designated category detection boxes, performance indicators, and custom training layout analysis models. For details, please refer to [document](layout/README.md). -Table Recognition converts table image into excel documents, which include the detection and recognition of table text and the prediction of table structure and cell coordinates. For detailed, please refer to [document](table/README.md) +#### 6.1.2 Table recognition -## 5. Prediction by inference engine +Table recognition converts table images into excel documents, which include the detection and recognition of table text and the prediction of table structure and cell coordinates. For detailed instructions, please refer to [document](table/README.md) -Use the following commands to complete the inference. + -```python -cd PaddleOCR/ppstructure +### 6.2 DOC-VQA -# download model -mkdir inference && cd inference -# Download the detection model of the ultra-lightweight Chinese OCR model and uncompress it -wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_det_infer.tar -# Download the recognition model of the ultra-lightweight Chinese OCR model and uncompress it -wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar -# Download the table structure model of the ultra-lightweight Chinese OCR model and uncompress it -wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar && tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar -cd .. +Document Visual Question Answering (DOC-VQA) if a type of Visual Question Answering (VQA), which includes Semantic Entity Recognition (SER) and Relation Extraction (RE) tasks. Based on SER task, text recognition and classification in images can be completed. Based on THE RE task, we can extract the relation of the text content in the image, such as judge the problem pair. For details, please refer to [document](vqa/README.md) -python3 predict_system.py --det_model_dir=inference/ch_ppocr_mobile_v2.0_det_infer --rec_model_dir=inference/ch_ppocr_mobile_v2.0_rec_infer --table_model_dir=inference/en_ppocr_mobile_v2.0_table_structure_infer --image_dir=../doc/table/1.png --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt --output=../output/table --vis_font_path=../doc/fonts/simfang.ttf -``` -After running, each image will have a directory with the same name under the directory specified in the output field. Each table in the picture will be stored as an excel and figure area will be cropped and saved, the excel and image file name will be the coordinates of the table in the image. -**Model List** + -|model name|description|config|model size|download| -| --- | --- | --- | --- | --- | -|en_ppocr_mobile_v2.0_table_structure|Table structure prediction for English table scenarios|[table_mv3.yml](../configs/table/table_mv3.yml)|18.6M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) | +## 7. Model List -**Model List** +PP-Structure系列模型列表（更新中） -LayoutParser model +* Layout analysis model |model name|description|download| | --- | --- | --- | -| ppyolov2_r50vd_dcn_365e_publaynet | The layout analysis model trained on the PubLayNet data set can be divided into 5 types of areas **text, title, table, picture and list** | [PubLayNet](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet.tar) | -| ppyolov2_r50vd_dcn_365e_tableBank_word | The layout analysis model trained on the TableBank Word dataset can only detect tables | [TableBank Word](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_word.tar) | -| ppyolov2_r50vd_dcn_365e_tableBank_latex | The layout analysis model trained on the TableBank Latex dataset can only detect tables | [TableBank Latex](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_latex.tar) | +| ppyolov2_r50vd_dcn_365e_publaynet | The layout analysis model trained on the PubLayNet dataset can divide image into 5 types of areas **text, title, table, picture, and list** | [PubLayNet](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet.tar) | -OCR and table recognition model + +* OCR and table recognition model |model name|description|model size|download| | --- | --- | --- | --- | |ch_ppocr_mobile_slim_v2.0_det|Slim pruned lightweight model, supporting Chinese, English, multilingual text detection|2.6M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_det_prune_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_det_prune_infer.tar) | |ch_ppocr_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting Chinese, English and number recognition|6M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_train.tar) | -|en_ppocr_mobile_v2.0_table_det|Text detection of English table scenes trained on PubLayNet dataset|4.7M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_det_train.tar) | -|en_ppocr_mobile_v2.0_table_rec|Text recognition of English table scene trained on PubLayNet dataset|6.9M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar) [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_rec_train.tar) | -|en_ppocr_mobile_v2.0_table_structure|Table structure prediction of English table scene trained on PubLayNet dataset|18.6M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) | +|en_ppocr_mobile_v2.0_table_structure|Table structure prediction of English table scene trained on PubLayNet dataset|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) | + +* DOC-VQA model +|model name|description|model size|download| +| --- | --- | --- | --- | +|PP-Layout_v1.0_ser_pretrained|SER model trained on xfun Chinese dataset based on LayoutXLM|1.4G|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_ser_pretrained.tar) | +|PP-Layout_v1.0_re_pretrained|RE model trained on xfun Chinese dataset based on LayoutXLM|1.4G|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_re_pretrained.tar) | -If you need to use other models, you can download the model in [model_list](../doc/doc_en/models_list_en.md) or use your own trained model to configure it to the three fields of `det_model_dir`, `rec_model_dir`, `table_model_dir` . +If you need to use other models, you can download the model in [PPOCR model_list](../doc/doc_en/models_list_en.md) and [PPStructure model_list](./docs/model_list.md) diff --git a/ppstructure/README_ch.md b/ppstructure/README_ch.md index 607efac1bf6bfaa58f0e96ceef1a0ee344189e9c..808a5c68d18df625bedeae4706da7f985d6caecd 100644 --- a/ppstructure/README_ch.md +++ b/ppstructure/README_ch.md @@ -1,14 +1,32 @@ [English](README.md) | 简体中文 -## 简介 +- [1. 简介](#1) +- [2. 近期更新](#2) +- [3. 特性](#3) +- [4. 效果展示](#4) + * [4.1 版面分析和表格识别](#41) + * [4.2 DOC-VQA](#42) +- [5. 快速体验](#5) +- [6. PP-Structure 介绍](#6) + * [6.1 版面分析+表格识别](#61) + * [6.2 DOC-VQA](#62) +- [7. 模型库](#7) + + + +## 1. 简介 PP-Structure是一个可用于复杂文档结构分析和处理的OCR工具包，旨在帮助开发者更好的完成文档理解相关任务。 -## 近期更新 -* 2021.12.07 新增VQA任务-SER和RE。 + -## 特性 +## 2. 近期更新 +* 2021.12.07 新增DOC-[VQA任务SER和RE](vqa/README.md)。 -PP-Structure是一个可用于复杂文档结构分析和处理的OCR工具包，主要特性如下： + + +## 3. 特性 + +PP-Structure的主要特性如下： - 支持对图片形式的文档进行版面分析，可以划分**文字、标题、表格、图片以及列表**5类区域（与Layout-Parser联合使用） - 支持文字、标题、图片以及列表区域提取为文字字段（与PP-OCR联合使用） - 支持表格区域进行结构化分析，最终结果输出Excel文件 @@ -17,13 +35,22 @@ PP-Structure是一个可用于复杂文档结构分析和处理的OCR工具包 - 支持文档视觉问答(Document Visual Question Answering，DOC-VQA)任务-语义实体识别(Semantic Entity Recognition，SER)和关系抽取(Relation Extraction，RE) -## 1. 效果展示 + + +## 4. 效果展示 -### 1.1 版面分析和表格识别 + + +### 4.1 版面分析和表格识别 -### 1.2 VQA +图中展示了版面分析+表格识别的整体流程，图片先有版面分析划分为图像、文本、标题和表格四种区域，然后对图像、文本和标题三种区域进行OCR的检测识别，对表格进行表格识别，其中图像还会被存储下来以便使用。 + + + + +### 4.2 DOC-VQA * SER @@ -46,36 +73,45 @@ PP-Structure是一个可用于复杂文档结构分析和处理的OCR工具包 图中红色框表示问题，蓝色框表示答案，问题和答案之间使用绿色线连接。在OCR检测框的左上方也标出了对应的类别和OCR识别结果。 -## 2. 快速体验 + + +## 5. 快速体验 + +请参考[快速安装](./docs/quickstart.md)教程。 -代码体验：从 [快速安装](./docs/quickstart.md) 开始 + -## 3. PP-Structure Pipeline介绍 +## 6. PP-Structure 介绍 -### 3.1 版面分析+表格识别 + + +### 6.1 版面分析+表格识别 ![pipeline](../doc/table/pipeline.jpg) 在PP-Structure中，图片会先经由Layout-Parser进行版面分析，在版面分析中，会对图片里的区域进行分类，包括**文字、标题、图片、列表和表格**5类。对于前4类区域，直接使用PP-OCR完成对应区域文字检测与识别。对于表格类区域，经过表格结构化处理后，表格图片转换为相同表格样式的Excel文件。 -#### 3.1.1 版面分析 +#### 6.1.1 版面分析 版面分析对文档数据进行区域分类，其中包括版面分析工具的Python脚本使用、提取指定类别检测框、性能指标以及自定义训练版面分析模型，详细内容可以参考[文档](layout/README_ch.md)。 -#### 3.1.2 表格识别 +#### 6.1.2 表格识别 + +表格识别将表格图片转换为excel文档，其中包含对于表格文本的检测和识别以及对于表格结构和单元格坐标的预测，详细说明参考[文档](table/README_ch.md)。 -表格识别将表格图片转换为excel文档，其中包含对于表格文本的检测和识别以及对于表格结构和单元格坐标的预测，详细说明参考[文档](table/README_ch.md) + +### 6.2 DOC-VQA -### 3.2 VQA +DOC-VQA指文档视觉问答，其中包括语义实体识别 (Semantic Entity Recognition, SER) 和关系抽取 (Relation Extraction, RE) 任务。基于 SER 任务，可以完成对图像中的文本识别与分类；基于 RE 任务，可以完成对图象中的文本内容的关系提取，如判断问题对(pair)，详细说明参考[文档](vqa/README.md)。 -coming soon + -## 4. 模型库 +## 7. 模型库 PP-Structure系列模型列表（更新中） -* LayoutParser 模型 +* 版面分析模型 |模型名称|模型简介|下载地址| | --- | --- | --- | @@ -90,7 +126,7 @@ PP-Structure系列模型列表（更新中） |ch_ppocr_mobile_slim_v2.0_rec|slim裁剪量化版超轻量模型，支持中英文、数字识别|6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_train.tar) | |en_ppocr_mobile_v2.0_table_structure|PubLayNet数据集训练的英文表格场景的表格结构预测|18.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) | -* VQA模型 +* DOC-VQA 模型 |模型名称|模型简介|模型大小|下载地址| | --- | --- | --- | --- | @@ -98,4 +134,4 @@ PP-Structure系列模型列表（更新中） |PP-Layout_v1.0_re_pretrained|基于LayoutXLM在xfun中文数据集上训练的RE模型|1.4G|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_re_pretrained.tar) | -更多模型下载，可以参考 [模型库](./docs/model_list.md) +更多模型下载，可以参考 [PPOCR model_list](../doc/doc_en/models_list.md) and [PPStructure model_list](./docs/model_list.md) \ No newline at end of file diff --git a/ppstructure/predict_system.py b/ppstructure/predict_system.py index e87499ccc410ae67a170f63301e5a99ef948b161..3f3dc65875a20b3f66403afecfd60f04e3d83d61 100644 --- a/ppstructure/predict_system.py +++ b/ppstructure/predict_system.py @@ -30,7 +30,6 @@ from ppocr.utils.utility import get_image_file_list, check_and_read_gif from ppocr.utils.logging import get_logger from tools.infer.predict_system import TextSystem from ppstructure.table.predict_table import TableSystem, to_excel -from ppstructure.vqa.infer_ser_e2e import SerPredictor, draw_ser_results from ppstructure.utility import parse_args, draw_structure_result logger = get_logger() @@ -66,6 +65,7 @@ class OCRSystem(object): self.use_angle_cls = args.use_angle_cls self.drop_score = args.drop_score elif self.mode == 'vqa': + from ppstructure.vqa.infer_ser_e2e import SerPredictor, draw_ser_results self.vqa_engine = SerPredictor(args) def __call__(self, img): diff --git a/ppstructure/vqa/README.md b/ppstructure/vqa/README.md index ce90d41beb13fb055092ecc8fd7b1b719b96d00e..619ada71a82eacd88abd39199d0b220dc6c64c9b 100644 --- a/ppstructure/vqa/README.md +++ b/ppstructure/vqa/README.md @@ -61,7 +61,7 @@ PP-Structure 里的 DOC-VQA算法基于PaddleNLP自然语言处理算法库进 - **（1) 安装PaddlePaddle** ```bash -pip3 install --upgrade pip +python3 -m pip install --upgrade pip # GPU安装 python3 -m pip install paddlepaddle-gpu==2.2 -i https://mirror.baidu.com/pypi/simple @@ -78,7 +78,7 @@ python3 -m pip install paddlepaddle==2.2 -i https://mirror.baidu.com/pypi/simple - **（1）pip快速安装PaddleOCR whl包（仅预测）** ```bash -pip install paddleocr +python3 -m pip install paddleocr ``` - **（2）下载VQA源码（预测+训练）** @@ -92,21 +92,11 @@ git clone https://gitee.com/paddlepaddle/PaddleOCR # 注：码云托管代码可能无法实时同步本github项目更新，存在3~5天延时，请优先使用推荐方式。 ``` -- **（3）安装PaddleNLP** - -```bash -# 需要使用PaddleNLP最新的代码版本进行安装 -git clone https://github.com/PaddlePaddle/PaddleNLP -b develop -cd PaddleNLP -pip3 install -e . -``` - - -- **（4）安装VQA的`requirements`** +- **（3）安装VQA的`requirements`** ```bash cd ppstructure/vqa -pip install -r requirements.txt +python3 -m pip install -r requirements.txt ``` ## 4. 使用 @@ -133,7 +123,7 @@ wget https://paddleocr.bj.bcebos.com/dataset/XFUND.tar * 启动训练 ```shell -python3.7 train_ser.py \ +python3 train_ser.py \ --model_name_or_path "layoutxlm-base-uncased" \ --ser_model_type "LayoutXLM" \ --train_data_dir "XFUND/zh_train/image" \ @@ -154,7 +144,7 @@ python3.7 train_ser.py \ * 恢复训练 ```shell -python3.7 train_ser.py \ +python3 train_ser.py \ --model_name_or_path "model_path" \ --ser_model_type "LayoutXLM" \ --train_data_dir "XFUND/zh_train/image" \ @@ -191,7 +181,7 @@ python3 eval_ser.py \ ```shell export CUDA_VISIBLE_DEVICES=0 -python3.7 infer_ser.py \ +python3 infer_ser.py \ --model_name_or_path "PP-Layout_v1.0_ser_pretrained/" \ --ser_model_type "LayoutXLM" \ --output_dir "output/ser/" \ @@ -205,7 +195,7 @@ python3.7 infer_ser.py \ ```shell export CUDA_VISIBLE_DEVICES=0 -python3.7 infer_ser_e2e.py \ +python3 infer_ser_e2e.py \ --model_name_or_path "PP-Layout_v1.0_ser_pretrained/" \ --ser_model_type "LayoutXLM" \ --max_seq_length 512 \ @@ -217,7 +207,7 @@ python3.7 infer_ser_e2e.py \ ```shell export CUDA_VISIBLE_DEVICES=0 -python3.7 helper/eval_with_label_end2end.py --gt_json_path XFUND/zh_val/xfun_normalize_val.json --pred_json_path output_res/infer_results.txt +python3 helper/eval_with_label_end2end.py --gt_json_path XFUND/zh_val/xfun_normalize_val.json --pred_json_path output_res/infer_results.txt ``` @@ -312,7 +302,7 @@ python3 infer_re.py \ ```shell export CUDA_VISIBLE_DEVICES=0 -python3.7 infer_ser_re_e2e.py \ +python3 infer_ser_re_e2e.py \ --model_name_or_path "PP-Layout_v1.0_ser_pretrained/" \ --re_model_name_or_path "PP-Layout_v1.0_re_pretrained/" \ --ser_model_type "LayoutXLM" \ diff --git a/ppstructure/vqa/infer.sh b/ppstructure/vqa/infer.sh deleted file mode 100644 index 2cd1cea4476672732b3a7f9ad97a3e42172dbb92..0000000000000000000000000000000000000000 --- a/ppstructure/vqa/infer.sh +++ /dev/null @@ -1,61 +0,0 @@ -export CUDA_VISIBLE_DEVICES=6 -# python3.7 infer_ser_e2e.py \ -# --model_name_or_path "output/ser_distributed/best_model" \ -# --max_seq_length 512 \ -# --output_dir "output_res_e2e/" \ -# --infer_imgs "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/zh.val/zh_val_0.jpg" - - -# python3.7 infer_ser_re_e2e.py \ -# --model_name_or_path "output/ser_distributed/best_model" \ -# --re_model_name_or_path "output/re_test/best_model" \ -# --max_seq_length 512 \ -# --output_dir "output_ser_re_e2e_train/" \ -# --infer_imgs "images/input/zh_val_21.jpg" - -# python3.7 infer_ser.py \ -# --model_name_or_path "output/ser_LayoutLM/best_model" \ -# --ser_model_type "LayoutLM" \ -# --output_dir "ser_LayoutLM/" \ -# --infer_imgs "images/input/zh_val_21.jpg" \ -# --ocr_json_path "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/xfun_normalize_val.json" - -python3.7 infer_ser.py \ - --model_name_or_path "output/ser_new/best_model" \ - --ser_model_type "LayoutXLM" \ - --output_dir "ser_new/" \ - --infer_imgs "images/input/zh_val_21.jpg" \ - --ocr_json_path "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/xfun_normalize_val.json" - -# python3.7 infer_ser_e2e.py \ -# --model_name_or_path "output/ser_new/best_model" \ -# --ser_model_type "LayoutXLM" \ -# --max_seq_length 512 \ -# --output_dir "output/ser_new/" \ -# --infer_imgs "images/input/zh_val_0.jpg" - - -# python3.7 infer_ser_e2e.py \ -# --model_name_or_path "output/ser_LayoutLM/best_model" \ -# --ser_model_type "LayoutLM" \ -# --max_seq_length 512 \ -# --output_dir "output/ser_LayoutLM/" \ -# --infer_imgs "images/input/zh_val_0.jpg" - -# python3 infer_re.py \ -# --model_name_or_path "/ssd1/zhoujun20/VQA/PaddleOCR/ppstructure/vqa/output/re_test/best_model/" \ -# --max_seq_length 512 \ -# --eval_data_dir "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/zh.val" \ -# --eval_label_path "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/xfun_normalize_val.json" \ -# --label_map_path 'labels/labels_ser.txt' \ -# --output_dir "output_res" \ -# --per_gpu_eval_batch_size 1 \ -# --seed 2048 - -# python3.7 infer_ser_re_e2e.py \ -# --model_name_or_path "output/ser_LayoutLM/best_model" \ -# --ser_model_type "LayoutLM" \ -# --re_model_name_or_path "output/re_new/best_model" \ -# --max_seq_length 512 \ -# --output_dir "output_ser_re_e2e/" \ -# --infer_imgs "images/input/zh_val_21.jpg" \ No newline at end of file diff --git a/ppstructure/vqa/requirements.txt b/ppstructure/vqa/requirements.txt index 9c935ae619024c9f47ced820eae35a3a1c976953..0042ec0baedcc3e7bbecb922d10b93c95219219d 100644 --- a/ppstructure/vqa/requirements.txt +++ b/ppstructure/vqa/requirements.txt @@ -1,3 +1,4 @@ sentencepiece yacs -seqeval \ No newline at end of file +seqeval +paddlenlp>=2.2.1 \ No newline at end of file