chinese-ocr
说明： keras/pytorch实现crnn+ctc实现不定长中文OCR识别
文字方向检测 0、90、180、270度检测
文字检测 后期将切换到keras版本文本检测 实现keras端到端的文本检测及识别
不定长OCR识别
(Kernas/pytorch Implementation of crnn+ctc for Uncertain Length Chinese OCR Recognition Text Direction Detection 0,90,180,270 Degree Detection Text detection will switch to keras version in the later stage of text detection to realize keras end-to-end text detection and recognition Uncertain Length OCR Recognition)
(Kernas/pytorch Implementation of crnn+ctc for Uncertain Length Chinese OCR Recognition Text Direction Detection 0,90,180,270 Degree Detection Text detection will switch to keras version in the later stage of text detection to realize keras end-to-end text detection and recognition Uncertain Length OCR Recognition)
文件列表:
ocr\keys.py (16429, 2019-03-22)
ocr\keys.pyc (16513, 2019-03-22)
ocr\model.py (4448, 2019-03-22)
ocr\model.pyc (4801, 2019-03-22)
ocr\__init__.py (0, 2019-03-22)
ocr\__init__.pyc (102, 2019-03-22)
test\.DS_Store (6148, 2019-03-22)
train\create-dataset.sh (45, 2019-03-22)
train\create_dataset\create_dataset.py (3326, 2019-03-22)
train\data\.DS_Store (6148, 2019-03-22)
train\data\dataline\ff299a9c-b41b-11e7-89e1-1c1b0d6ddf51.jpg (3860, 2019-03-22)
train\data\dataline\ff299a9c-b41b-11e7-89e1-1c1b0d6ddf51.txt (30, 2019-03-22)
train\data\lmdb\.DS_Store (6148, 2019-03-22)
train\data\lmdb\train\data.mdb (319488, 2019-03-22)
train\data\lmdb\train\lock.mdb (8192, 2019-03-22)
train\data\lmdb\val\data.mdb (319488, 2019-03-22)
train\data\lmdb\val\lock.mdb (8192, 2019-03-22)
train\keras-train\.DS_Store (6148, 2019-03-22)
train\keras-train\dataset.py (4202, 2019-03-22)
train\keras-train\dataset.pyc (5621, 2019-03-22)
train\keras-train\keys.py (16429, 2019-03-22)
train\keras-train\keys.pyc (16561, 2019-03-22)
train\keras-train\model.py (2893, 2019-03-22)
train\keras-train\model.pyc (3573, 2019-03-22)
train\keras-train\train.py (2925, 2019-03-22)
train\keras-train\train.pyc (3100, 2019-03-22)
train\keras-train\trainbatch.py (1958, 2019-03-22)
train\pretrain-models\.DS_Store (6148, 2019-03-22)
train\pytorch-train\.DS_Store (6148, 2019-03-22)
train\pytorch-train\crnn_main.py (10046, 2019-03-22)
train\pytorch-train\crnn_main.py~ (10039, 2019-03-22)
train\pytorch-train\dataset.py (4101, 2019-03-22)
train\pytorch-train\dataset.pyc (5524, 2019-03-22)
train\pytorch-train\keys.py (16423, 2019-03-22)
train\pytorch-train\keys.pyc (16556, 2019-03-22)
train\pytorch-train\log.log (294727, 2019-03-22)
train\pytorch-train\models\crnn.py (3102, 2019-03-22)
train\pytorch-train\models\crnn.pyc (3688, 2019-03-22)
train\pytorch-train\models\utils.py (334, 2019-03-22)
train\pytorch-train\models\utils.pyc (620, 2019-03-22)
... ...
ocr\keys.pyc (16513, 2019-03-22)
ocr\model.py (4448, 2019-03-22)
ocr\model.pyc (4801, 2019-03-22)
ocr\__init__.py (0, 2019-03-22)
ocr\__init__.pyc (102, 2019-03-22)
test\.DS_Store (6148, 2019-03-22)
train\create-dataset.sh (45, 2019-03-22)
train\create_dataset\create_dataset.py (3326, 2019-03-22)
train\data\.DS_Store (6148, 2019-03-22)
train\data\dataline\ff299a9c-b41b-11e7-89e1-1c1b0d6ddf51.jpg (3860, 2019-03-22)
train\data\dataline\ff299a9c-b41b-11e7-89e1-1c1b0d6ddf51.txt (30, 2019-03-22)
train\data\lmdb\.DS_Store (6148, 2019-03-22)
train\data\lmdb\train\data.mdb (319488, 2019-03-22)
train\data\lmdb\train\lock.mdb (8192, 2019-03-22)
train\data\lmdb\val\data.mdb (319488, 2019-03-22)
train\data\lmdb\val\lock.mdb (8192, 2019-03-22)
train\keras-train\.DS_Store (6148, 2019-03-22)
train\keras-train\dataset.py (4202, 2019-03-22)
train\keras-train\dataset.pyc (5621, 2019-03-22)
train\keras-train\keys.py (16429, 2019-03-22)
train\keras-train\keys.pyc (16561, 2019-03-22)
train\keras-train\model.py (2893, 2019-03-22)
train\keras-train\model.pyc (3573, 2019-03-22)
train\keras-train\train.py (2925, 2019-03-22)
train\keras-train\train.pyc (3100, 2019-03-22)
train\keras-train\trainbatch.py (1958, 2019-03-22)
train\pretrain-models\.DS_Store (6148, 2019-03-22)
train\pytorch-train\.DS_Store (6148, 2019-03-22)
train\pytorch-train\crnn_main.py (10046, 2019-03-22)
train\pytorch-train\crnn_main.py~ (10039, 2019-03-22)
train\pytorch-train\dataset.py (4101, 2019-03-22)
train\pytorch-train\dataset.pyc (5524, 2019-03-22)
train\pytorch-train\keys.py (16423, 2019-03-22)
train\pytorch-train\keys.pyc (16556, 2019-03-22)
train\pytorch-train\log.log (294727, 2019-03-22)
train\pytorch-train\models\crnn.py (3102, 2019-03-22)
train\pytorch-train\models\crnn.pyc (3688, 2019-03-22)
train\pytorch-train\models\utils.py (334, 2019-03-22)
train\pytorch-train\models\utils.pyc (620, 2019-03-22)
... ...
# 本文基于tensorflow、keras/pytorch实现对自然场景的文字检测及端到端的OCR中文文字识别
# 实现功能
- [x] 文字方向检测 0、90、180、270度检测
- [x] 文字检测 后期将切换到keras版本文本检测 实现keras端到端的文本检测及识别
- [x] 不定长OCR识别
## 环境部署
``` Bash
##GPU环境
sh setup.sh
##CPU环境
sh setup-cpu.sh
```
# 模型训练
## 训练keras版本的crnn
``` Bash
cd train & sh train-keras.sh
```
## 训练pytorch版本的crnn
``` Bash
cd train & sh train-pytorch.sh
```
# 文字方向检测
基于图像分类,在VGG16模型的基础上,迁移训练0、90、180、270度的文字方向分类模型,详细代码参考angle/predict.py文件,训练图片100000张,准确率95.10%。
模型地址[百度云](https://pan.baidu.com/s/1nwEyxDZ)下载
# 文字检测
支持CPU、GPU环境,一键部署,
[文本检测训练参考](https://github.com/eragonruan/text-detection-ctpn)(https://github.com/eragonruan/text-detection-ctpn)
# OCR 端到端识别:GRU+CTC
## ocr识别采用GRU+CTC端到到识别技术,实现不分隔识别不定长文字
提供keras 与pytorch版本的训练代码,在理解keras的基础上,可以切换到pytorch版本,此版本更稳定
# 识别结果展示
## 文字检测及OCR识别结果
### 倾斜文字
## 参考
1.crnn https://github.com/meijieru/crnn.pytorch.git
2.keras-crnn 版本实现参考 https://www.zhihu.com/question/59***5822
3.ctpn https://github.com/eragonruan/text-detection-ctpn , https://github.com/tianzhi0549/CTPN
近期下载者：
相关文件：
收藏者：