PaddleOCR/doc/doc_ch/algorithm_rec_parseq.md

125 lines
4.1 KiB
Markdown
Raw Normal View History

Add new recognition method "ParseQ" (#10836) * Update PP-OCRv4_introduction.md * Update PP-OCRv4_introduction.md (#10616) * Update PP-OCRv4_introduction.md * Update PP-OCRv4_introduction.md * Update PP-OCRv4_introduction.md * Update README.md * Cherrypicking GH-10217 and GH-10216 to PaddlePaddle:Release/2.7 (#10655) * Don't break overall processing on a bad image * Add preprocessing common to OCR tasks Add preprocessing to options * Update requirements.txt (#10656) added missing pyyaml library * [TIPC]update xpu tipc script (#10658) * fix-typo (#10642) Co-authored-by: Dennis <dvorst@users.noreply.github.com> Co-authored-by: shiyutang <34859558+shiyutang@users.noreply.github.com> * 修改数据增强导致的DSR报错 (#10662) (#10681) * 修改数据增强导致的DSR报错 * 错误修改回滚 * Update algorithm_overview_en.md (#10670) Fixed simple spelling errors. * Implement recoginition method ParseQ * Document update for new recognition method ParseQ * add prediction for parseq * Update rec_vit_parseq.yml * Update rec_r31_sar.yml * Update rec_r31_sar.yml * Update rec_r50_fpn_srn.yml * Update rec_vit_parseq.py * Update rec_vit_parseq.yml * Update rec_parseq_head.py * Update rec_img_aug.py * Update rec_vit_parseq.yml * Update __init__.py * Update predict_rec.py * Update paddleocr.py * Update requirements.txt * Update utility.py * Update utility.py --------- Co-authored-by: xiaoting <31891223+tink2123@users.noreply.github.com> Co-authored-by: topduke <784990967@qq.com> Co-authored-by: dyning <dyning.2003@163.com> Co-authored-by: UserUnknownFactor <63057995+UserUnknownFactor@users.noreply.github.com> Co-authored-by: itasli <ilyas.tasli@outlook.fr> Co-authored-by: Kai Song <50285351+USTCKAY@users.noreply.github.com> Co-authored-by: dvorst <87502756+dvorst@users.noreply.github.com> Co-authored-by: Dennis <dvorst@users.noreply.github.com> Co-authored-by: shiyutang <34859558+shiyutang@users.noreply.github.com> Co-authored-by: Dec20B <1192152456@qq.com> Co-authored-by: ncoffman <51147417+ncoffman@users.noreply.github.com>
2023-09-07 16:36:47 +08:00
# ParseQ
- [1. 算法简介](#1)
- [2. 环境配置](#2)
- [3. 模型训练、评估、预测](#3)
- [3.1 训练](#3-1)
- [3.2 评估](#3-2)
- [3.3 预测](#3-3)
- [4. 推理部署](#4)
- [4.1 Python推理](#4-1)
- [4.2 C++推理](#4-2)
- [4.3 Serving服务化部署](#4-3)
- [4.4 更多推理部署](#4-4)
- [5. FAQ](#5)
<a name="1"></a>
## 1. 算法简介
论文信息:
> [Scene Text Recognition with Permuted Autoregressive Sequence Models](https://arxiv.org/abs/2207.06966)
> Darwin Bautista, Rowel Atienza
> ECCV, 2021
原论文分别使用真实文本识别数据集(Real)和合成文本识别数据集(Synth)进行训练在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估。
其中:
- 真实文本识别数据集(Real)包含COCO-Text, RCTW17, Uber-Text, ArT, LSVT, MLT19, ReCTS, TextOCR, OpenVINO数据集
- 合成文本识别数据集(Synth)包含MJSynth和SynthText数据集
在不同数据集上训练的算法的复现效果如下:
|数据集|模型|骨干网络|配置文件|Acc|下载链接|
| --- | --- | --- | --- | --- | --- |
|Synth|ParseQ|VIT|[rec_vit_parseq.yml](../../configs/rec/rec_vit_parseq.yml)|91.24%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_synth.tgz)|
|Real|ParseQ|VIT|[rec_vit_parseq.yml](../../configs/rec/rec_vit_parseq.yml)|94.74%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_real.tgz)|
|||||||
<a name="2"></a>
## 2. 环境配置
请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境参考[《项目克隆》](./clone.md)克隆项目代码。
<a name="3"></a>
## 3. 模型训练、评估、预测
请参考[文本识别教程](./recognition.md)。PaddleOCR对代码进行了模块化训练不同的识别模型只需要**更换配置文件**即可。
训练
具体地,在完成数据准备后,便可以启动训练,训练命令如下:
```
#单卡训练(训练周期长,不建议)
python3 tools/train.py -c configs/rec/rec_vit_parseq.yml
#多卡训练,通过--gpus参数指定卡号
python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_vit_parseq.yml
```
评估
```
# GPU 评估, Global.pretrained_model 为待测权重
python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
```
预测:
```
# 预测使用的配置文件必须与训练一致
python3 tools/infer_rec.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
```
<a name="4"></a>
## 4. 推理部署
<a name="4-1"></a>
### 4.1 Python推理
首先将ParseQ文本识别训练过程中保存的模型转换成inference model。 [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_real.tgz) ),可以使用如下命令进行转换:
```
python3 tools/export_model.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model=./rec_vit_parseq_real/best_accuracy Global.save_inference_dir=./inference/rec_parseq
```
ParseQ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_1.png" --rec_model_dir="./inference/rec_parseq/" --rec_image_shape="3, 32, 128" --rec_algorithm="ParseQ" --rec_char_dict_path="ppocr/utils/dict/parseq_dict.txt" --max_text_length=25 --use_space_char=False
```
<a name="4-2"></a>
### 4.2 C++推理
由于C++预处理后处理还未支持ParseQ所以暂未支持
<a name="4-3"></a>
### 4.3 Serving服务化部署
暂不支持
<a name="4-4"></a>
### 4.4 更多推理部署
暂不支持
<a name="5"></a>
## 5. FAQ
## 引用
```bibtex
@InProceedings{bautista2022parseq,
title={Scene Text Recognition with Permuted Autoregressive Sequence Models},
author={Bautista, Darwin and Atienza, Rowel},
booktitle={European Conference on Computer Vision},
pages={178--196},
month={10},
year={2022},
publisher={Springer Nature Switzerland},
address={Cham},
doi={10.1007/978-3-031-19815-1_11},
url={https://doi.org/10.1007/978-3-031-19815-1_11}
}
```