Skip to content

GitLab

"magic_pdf/git@developer.sourcefind.cn:wangsen/mineru.git" did not exist on "196de029a3d2ef933699912b90233ca4b20d1bcd"
Unverified Commit 4e87349b authored by zhoujun's avatar zhoujun Committed by GitHub
Browse files

Merge pull request #1431 from WenmuZhou/tree_doc 

[Dygraph] add  Algorithm overview
parents 1f926f5f 74c70afa
Hide whitespace changes
Inline Side-by-side
Showing with 52 additions and 73 deletions
doc/doc_ch/algorithm_overview.md
View file @ 4e87349b
<a name="算法介绍"></a>
## 算法介绍
本文给出了PaddleOCR已支持的文本检测算法和文本识别算法列表，以及每个算法在**英文公开数据集**上的模型和指标，主要用于算法简介和算法性能对比，更多包括中文在内的其他数据集上的模型请参考[PP-OCR v1.1 系列模型下载](./models_list.md)。
本文给出了PaddleOCR已支持的文本检测算法和文本识别算法列表，以及每个算法在**英文公开数据集**上的模型和指标，主要用于算法简介和算法性能对比，更多包括中文在内的其他数据集上的模型请参考[PP-OCR v2.0 系列模型下载](./models_list.md)。
- [1.文本检测算法](#文本检测算法)
- [2.文本识别算法](#文本识别算法)
......@@ -9,25 +9,25 @@
### 1.文本检测算法
PaddleOCR开源的文本检测算法列表：
- [x] DB([paper](https://arxiv.org/abs/1911.08947))（ppocr推荐）
- [x] DB([paper]( https://arxiv.org/abs/1911.08947) )（ppocr推荐）
- [x] EAST([paper](https://arxiv.org/abs/1704.03155))
- [x] SAST([paper](https://arxiv.org/abs/1908.05498))
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|88.18%|85.51%|86.82%|[下载链接](link)|
|EAST|MobileNetV3|81.67%|79.83%|80.74%|[下载链接](link)|
|DB|ResNet50_vd|83.79%|80.65%|82.19%|[下载链接](link)|
|DB|MobileNetV3|75.92%|73.18%|74.53%|[下载链接](link)|
|SAST|ResNet50_vd|92.18%|82.96%|87.33%|[下载链接](link))|
| --- | --- | --- | --- | --- | --- |
|EAST|ResNet50_vd|88.18%|85.51%|86.82%|[下载链接 (coming soon)](link)|
|EAST|MobileNetV3|81.67%|79.83%|80.74%|[下载链接 (coming soon)](coming soon)|
|DB|ResNet50_vd|83.79%|80.65%|82.19%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
|DB|MobileNetV3|75.92%|73.18%|74.53%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
|SAST|ResNet50_vd|92.18%|82.96%|87.33%|[下载链接 (coming soon)](link)|
在Total-text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|
|-|-|-|-|-|-|
|SAST|ResNet50_vd|88.74%|79.80%|84.03%|[下载链接](link)|
| --- | --- | --- | --- | --- | --- |
|SAST|ResNet50_vd|88.74%|79.80%|84.03%|[下载链接 (coming soon)](link)|
**说明：** SAST模型训练额外加入了icdar2013、icdar2017、COCO-Text、ArT等公开数据集进行调优。PaddleOCR用到的经过整理格式的英文公开数据集下载：[百度云地址](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (提取码: 2bpi)
......@@ -38,22 +38,22 @@ PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训
### 2.文本识别算法
PaddleOCR基于动态图开源的文本识别算法列表：
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))（ppocr推荐）
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717) )（ppocr推荐）
- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))
- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [ ] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1)) coming soon
- [ ] SRN([paper](https://arxiv.org/abs/2003.12294)) coming soon
参考[DTRB](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
|模型|骨干网络|Avg Accuracy|模型存储命名|下载链接|
|-|-|-|-|-|
|Rosetta|Resnet34_vd|80.24%|rec_r34_vd_none_none_ctc|[下载链接](link)|
|Rosetta|MobileNetV3|78.16%|rec_mv3_none_none_ctc|[下载链接](link)|
|CRNN|Resnet34_vd|82.20%|rec_r34_vd_none_bilstm_ctc|[下载链接](link)|
|CRNN|MobileNetV3|79.37%|rec_mv3_none_bilstm_ctc|[下载链接](link)|
|STAR-Net|Resnet34_vd|83.93%|rec_r34_vd_tps_bilstm_ctc|[下载链接](link)|
|STAR-Net|MobileNetV3|81.56%|rec_mv3_tps_bilstm_ctc|[下载链接](link)|
| --- | --- | --- | --- | --- |
|Rosetta|MobileNetV3|78.05%|rec_mv3_none_none_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
|Rosetta|Resnet34_vd|80.9%|rec_r34_vd_none_none_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
|CRNN|MobileNetV3|79.97%|rec_mv3_none_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
|CRNN|Resnet34_vd|82.76%|rec_r34_vd_none_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
|STAR-Net|MobileNetV3|81.56%|rec_mv3_tps_bilstm_ctc|[下载链接 (coming soon )]()|
|STAR-Net|Resnet34_vd|83.93%|rec_r34_vd_tps_bilstm_ctc|[下载链接 (coming soon )]()|
PaddleOCR文本识别算法的训练和使用请参考文档教程中[模型训练/评估中的文本识别部分](./recognition.md)。
doc/doc_ch/angle_class.md
View file @ 4e87349b
......@@ -62,9 +62,9 @@ PaddleOCR提供了训练脚本、评估脚本和预测脚本。
*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
```
# GPU训练 支持单卡，多卡训练，通过selected_gpus指定卡号
# GPU训练 支持单卡，多卡训练，通过 '--gpus' 指定卡号，如果使用的paddle版本小于2.0rc1，请使用'--select_gpus'参数选择要使用的GPU
# 启动训练，下面的命令已经写入train.sh文件中，只需修改文件里的配置文件路径即可
python3 -m paddle.distributed.launch --selected_gpus '0,1,2,3,4,5,6,7' tools/train.py -c configs/cls/cls_mv3.yml
python3 -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7' tools/train.py -c configs/cls/cls_mv3.yml
```
- 数据增强
......@@ -74,7 +74,7 @@ PaddleOCR提供了多种数据增强方式，如果您希望在训练时加入
默认的扰动方式有：颜色空间转换(cvtColor)、模糊(blur)、抖动(jitter)、噪声(Gasuss noise)、随机切割(random crop)、透视(perspective)、颜色反转(reverse),随机数据增强(RandAugment)。
训练过程中除随机数据增强外每种扰动方式以50%的概率被选择，具体代码实现请参考：
[rec_img_aug.py](../../ppocr/data/imaug/rec_img_aug.py)
[rec_img_aug.py](../../ppocr/data/imaug/rec_img_aug.py)
[randaugment.py](../../ppocr/data/imaug/randaugment.py)
*由于OpenCV的兼容性问题，扰动操作暂时只支持linux*
......
doc/doc_ch/inference.md
View file @ 4e87349b
......@@ -22,9 +22,8 @@ inference 模型（`paddle.jit.save`保存的模型）
- [三、文本识别模型推理](#文本识别模型推理)
- [1. 超轻量中文识别模型推理](#超轻量中文识别模型推理)
- [2. 基于CTC损失的识别模型推理](#基于CTC损失的识别模型推理)
- [3. 基于Attention损失的识别模型推理](#基于Attention损失的识别模型推理)
- [4. 自定义文本识别字典的推理](#自定义文本识别字典的推理)
- [5. 多语言模型的推理](#多语言模型的推理)
- [3. 自定义文本识别字典的推理](#自定义文本识别字典的推理)
- [4. 多语言模型的推理](#多语言模型的推理)
- [四、方向分类模型推理](#方向识别模型推理)
- [1. 方向分类模型推理](#方向分类模型推理)
......@@ -268,16 +267,6 @@ CRNN 文本识别模型推理，可以执行如下命令：
python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_crnn/" --rec_image_shape="3, 32, 100" --rec_char_type="en"
```
<a name="基于Attention损失的识别模型推理"></a>
### 3. 基于Attention损失的识别模型推理
基于Attention损失的识别模型与ctc不同，需要额外设置识别算法参数 --rec_algorithm="RARE"
RARE 文本识别模型推理，可以执行如下命令：
```
python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rare/" --rec_image_shape="3, 32, 100" --rec_char_type="en" --rec_algorithm="RARE"
```
![](../imgs_words_en/word_336.png)
执行命令后，上面图像的识别结果如下：
......@@ -297,7 +286,7 @@ self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
dict_character = list(self.character_str)
```
### 4. 自定义文本识别字典的推理
### 3. 自定义文本识别字典的推理
如果训练时修改了文本的字典，在使用inference模型预测时，需要通过`--rec_char_dict_path`指定使用的字典路径，并且设置 `rec_char_type=ch`
```
......@@ -305,7 +294,7 @@ python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png
```
<a name="多语言模型的推理"></a>
### 5. 多语言模型的推理
### 4. 多语言模型的推理
如果您需要预测的是其他语言模型，在使用inference模型预测时，需要通过`--rec_char_dict_path`指定使用的字典路径, 同时为了得到正确的可视化结果，
需要通过 `--vis_font_path` 指定可视化的字体路径，`doc/` 路径下有默认提供的小语种字体，例如韩文识别：
......
doc/doc_en/algorithm_overview_en.md
View file @ 4e87349b
......@@ -13,23 +13,23 @@ This tutorial lists the text detection algorithms and text recognition algorithm
PaddleOCR open source text detection algorithms list:
- [x] EAST([paper](https://arxiv.org/abs/1704.03155))
- [x] DB([paper](https://arxiv.org/abs/1911.08947))
- [x] SAST([paper](https://arxiv.org/abs/1908.05498))(Baidu Self-Research)
- [x] SAST([paper](https://arxiv.org/abs/1908.05498) )(Baidu Self-Research)
On the ICDAR2015 dataset, the text detection result is as follows:
|Model|Backbone|precision|recall|Hmean|Download link|
|-|-|-|-|-|-|
|EAST|ResNet50_vd|88.18%|85.51%|86.82%|[Download link](link)|
|EAST|MobileNetV3|81.67%|79.83%|80.74%|[Download link](link)|
|DB|ResNet50_vd|83.79%|80.65%|82.19%|[Download link](link)|
|DB|MobileNetV3|75.92%|73.18%|74.53%|[Download link](link)|
|SAST|ResNet50_vd|92.18%|82.96%|87.33%|[Download link](link)|
| --- | --- | --- | --- | --- | --- |
|EAST|ResNet50_vd|88.18%|85.51%|86.82%|[download link (coming soon)](link)|
|EAST|MobileNetV3|81.67%|79.83%|80.74%|[download link (coming soon)](coming soon)|
|DB|ResNet50_vd|83.79%|80.65%|82.19%|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
|DB|MobileNetV3|75.92%|73.18%|74.53%|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
|SAST|ResNet50_vd|92.18%|82.96%|87.33%|[download link (coming soon)](link)|
On Total-Text dataset, the text detection result is as follows:
|Model|Backbone|precision|recall|Hmean|Download link|
|-|-|-|-|-|-|
|SAST|ResNet50_vd|88.74%|79.80%|84.03%|[Download link](link)|
| --- | --- | --- | --- | --- | --- |
|SAST|ResNet50_vd|88.74%|79.80%|84.03%|[download link (coming soon)](link)|
**Note：** Additional data, like icdar2013, icdar2017, COCO-Text, ArT, was added to the model training of SAST. Download English public dataset in organized format used by PaddleOCR from [Baidu Drive](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (download code: 2bpi).
......@@ -41,20 +41,21 @@ For the training guide and use of PaddleOCR text detection algorithms, please re
PaddleOCR open-source text recognition algorithms list:
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))
- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))
- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [ ] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1)) coming soon
- [ ] SRN([paper](https://arxiv.org/abs/2003.12294))(Baidu Self-Research) coming soon
- [ ] SRN([paper](https://arxiv.org/abs/2003.12294) )(Baidu Self-Research) coming soon
Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
|Model|Backbone|Avg Accuracy|Module combination|Download link|
|-|-|-|-|-|
|Rosetta|Resnet34_vd|80.24%|rec_r34_vd_none_none_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_none_none_ctc.tar)|
|Rosetta|MobileNetV3|78.16%|rec_mv3_none_none_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_none_none_ctc.tar)|
|CRNN|Resnet34_vd|82.20%|rec_r34_vd_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_none_bilstm_ctc.tar)|
|CRNN|MobileNetV3|79.37%|rec_mv3_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_none_bilstm_ctc.tar)|
|STAR-Net|Resnet34_vd|83.93%|rec_r34_vd_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_ctc.tar)|
|STAR-Net|MobileNetV3|81.56%|rec_mv3_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_ctc.tar)|
| --- | --- | --- | --- | --- |
|Rosetta|MobileNetV3|78.05%|rec_mv3_none_none_ctc|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
|Rosetta|Resnet34_vd|80.9%|rec_r34_vd_none_none_ctc|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
|CRNN|MobileNetV3|79.97%|rec_mv3_none_bilstm_ctc|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
|CRNN|Resnet34_vd|82.76%|rec_r34_vd_none_bilstm_ctc|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
|STAR-Net|MobileNetV3|81.56%|rec_mv3_tps_bilstm_ctc|[download link (coming soon )]()|
|STAR-Net|Resnet34_vd|83.93%|rec_r34_vd_tps_bilstm_ctc|[download link (coming soon )]()|
Please refer to the document for training guide and use of PaddleOCR text recognition algorithms [Text recognition model training/evaluation/prediction](./doc/doc_en/recognition_en.md)
doc/doc_en/angle_class_en.md
View file @ 4e87349b
......@@ -65,9 +65,9 @@ Start training:
```
# Set PYTHONPATH path
export PYTHONPATH=$PYTHONPATH:.
# GPU training Support single card and multi-card training, specify the card number through selected_gpus
# GPU training Support single card and multi-card training, specify the card number through --gpus. If your paddle version is less than 2.0rc1, please use '--selected_gpus'
# Start training, the following command has been written into the train.sh file, just modify the configuration file path in the file
python3 -m paddle.distributed.launch --selected_gpus '0,1,2,3,4,5,6,7' tools/train.py -c configs/cls/cls_mv3.yml
python3 -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7' tools/train.py -c configs/cls/cls_mv3.yml
```
- Data Augmentation
......@@ -77,7 +77,7 @@ PaddleOCR provides a variety of data augmentation methods. If you want to add di
The default perturbation methods are: cvtColor, blur, jitter, Gasuss noise, random crop, perspective, color reverse, RandAugment.
Except for RandAugment, each disturbance method is selected with a 50% probability during the training process. For specific code implementation, please refer to:
[rec_img_aug.py](../../ppocr/data/imaug/rec_img_aug.py)
[rec_img_aug.py](../../ppocr/data/imaug/rec_img_aug.py)
[randaugment.py](../../ppocr/data/imaug/randaugment.py)
......
doc/doc_en/inference_en.md
View file @ 4e87349b
......@@ -25,9 +25,8 @@ Next, we first introduce how to convert a trained model into an inference model,
- [TEXT RECOGNITION MODEL INFERENCE](#RECOGNITION_MODEL_INFERENCE)
- [1. LIGHTWEIGHT CHINESE MODEL](#LIGHTWEIGHT_RECOGNITION)
- [2. CTC-BASED TEXT RECOGNITION MODEL INFERENCE](#CTC-BASED_RECOGNITION)
- [3. ATTENTION-BASED TEXT RECOGNITION MODEL INFERENCE](#ATTENTION-BASED_RECOGNITION)
- [4. TEXT RECOGNITION MODEL INFERENCE USING CUSTOM CHARACTERS DICTIONARY](#USING_CUSTOM_CHARACTERS)
- [5. MULTILINGUAL MODEL INFERENCE](MULTILINGUAL_MODEL_INFERENCE)
- [3. TEXT RECOGNITION MODEL INFERENCE USING CUSTOM CHARACTERS DICTIONARY](#USING_CUSTOM_CHARACTERS)
- [4. MULTILINGUAL MODEL INFERENCE](MULTILINGUAL_MODEL_INFERENCE)
- [ANGLE CLASSIFICATION MODEL INFERENCE](#ANGLE_CLASS_MODEL_INFERENCE)
- [1. ANGLE CLASSIFICATION MODEL INFERENCE](#ANGLE_CLASS_MODEL_INFERENCE)
......@@ -275,15 +274,6 @@ For CRNN text recognition model inference, execute the following commands:
python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/starnet/" --rec_image_shape="3, 32, 100" --rec_char_type="en"
```
<a name="ATTENTION-BASED_RECOGNITION"></a>
### 3. ATTENTION-BASED TEXT RECOGNITION MODEL INFERENCE
The recognition model based on Attention loss is different from ctc, and additional recognition algorithm parameters need to be set --rec_algorithm="RARE"
After executing the command, the recognition result of the above image is as follows:
```bash
python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rare/" --rec_image_shape="3, 32, 100" --rec_char_type="en" --rec_algorithm="RARE"
```
![](../imgs_words_en/word_336.png)
After executing the command, the recognition result of the above image is as follows:
......@@ -303,7 +293,7 @@ dict_character = list(self.character_str)
```
<a name="USING_CUSTOM_CHARACTERS"></a>
### 4. TEXT RECOGNITION MODEL INFERENCE USING CUSTOM CHARACTERS DICTIONARY
### 3. TEXT RECOGNITION MODEL INFERENCE USING CUSTOM CHARACTERS DICTIONARY
If the text dictionary is modified during training, when using the inference model to predict, you need to specify the dictionary path used by `--rec_char_dict_path`, and set `rec_char_type=ch`
```
......@@ -311,7 +301,7 @@ python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png
```
<a name="MULTILINGUAL_MODEL_INFERENCE"></a>
### 5. MULTILINGAUL MODEL INFERENCE
### 4. MULTILINGAUL MODEL INFERENCE
If you need to predict other language models, when using inference model prediction, you need to specify the dictionary path used by `--rec_char_dict_path`. At the same time, in order to get the correct visualization results,
You need to specify the visual font path through `--vis_font_path`. There are small language fonts provided by default under the `doc/` path, such as Korean recognition:
......
ppocr/modeling/transforms/tps.py
View file @ 4e87349b
......@@ -180,7 +180,6 @@ class GridGenerator(nn.Layer):
P = self.build_P_paddle(I_r_size)
inv_delta_C_tensor = self.build_inv_delta_C_paddle(C).astype('float32')
# inv_delta_C_tensor = paddle.zeros((23,23)).astype('float32')
P_hat_tensor = self.build_P_hat_paddle(
C, paddle.to_tensor(P)).astype('float32')
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment