50 lines
1.2 KiB
Markdown
50 lines
1.2 KiB
Markdown
|
# onnxOCR
|
|||
|
|
#### 一.优势：
|
|||
|
|
1.脱离深度学习训练框架，可直接用于部署的通用OCR。
|
|||
|
|
2.在算力有限，精度不变的情况下使用paddleOCR转成ONNX模型，进行重新构建的一款可部署在arm架构和x86架构计算机上的OCR模型。
|
|||
|
|
3.在同样性能的计算机上推理速度加速了4-5倍。
|
|||
|
|
|
|||
|
|
#### 二.环境安装
|
|||
|
|
python>=3.6
|
|||
|
|
|
|||
|
|
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple -r requirements.txt
|
|||
|
|
|
|||
|
|
由于rec模型超过了100M，github有限制，所以我上传到
|
|||
|
|
[百度网盘,提取码: 125c](https://pan.baidu.com/s/1O1b30CMwsDjD7Ti9EnxYKQ )
|
|||
|
|
|
|||
|
|
下载后放到./models/ch_ppocr_server_v2.0/rec/rec.onnx下
|
|||
|
|
|
|||
|
|
#### 三.一键运行
|
|||
|
|
|
|||
|
|
python test_ocr.py
|
|||
|
|
|
|||
|
|
#### 效果展示
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
#### 感谢PaddleOcr
|
|||
|
|
|
|||
|
|
https://github.com/PaddlePaddle/PaddleOCR
|
|||
|
|
|
|||
|
|
#### 从该项目Fork而来
|
|||
|
|
https://github.com/jingsongliujing/OnnxOCR
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
CHANGELOG
|
|||
|
|
|
|||
|
|
1. 加入最新的`pp_ocr_v4`的检测与识别模型
|
|||
|
|
2. 修改包名为`pp_onnx`防止与onnx冲突
|
|||
|
|
3. 修改部分写死的参数
|
|||
|
|
|