E2E-MLT

代码fork自E2E-MLT，完整说明文档见Readme

Requirements

python3.x with
opencv-python
pytorch
torchvision

编译

NMS 模块

setuptool

Create file setup.py in nms directory and type the following script.

from setuptools import setup, Extension

functions_module = Extension(
    name='adaptor',
    sources=['adaptor.cpp', 'include/clipper/clipper.cpp'],
    #sources=['adaptor.cpp'],
    include_dirs=[r'.\include',
                  r'E:\install\Anaconda3\include'] # change to your path
)

setup(ext_modules=[functions_module])

Compile

First make sure c++ compiling toolchain is in PATH.

On windows its better to use VisualStudio's Prompt cmd env.

python setup.py build_ext --inplace

Warp CTC

Warp-ctc To make life easier, I just install the module with pip.

pip install torch-baidu-ctc

下载预训练模型

e2e-mlt, e2e-mlt-rctw

wget http://ptak.felk.cvut.cz/public_datasets/SyntText/e2e-mlt.h5

测试

下载的模型为 7500 字符集的，而代码中设置的为8400，需要改一下不然跑不起来。将 models.py 里面的 8400 替换为 7500，我已经改好了，直接用。

GPU 运行

python3 demo.py -model=e2e-mlt.h5

CPU 运行

跑起来比较慢，笔记本电脑跑起来非常的卡。

python3 demo.py -cuda=0 -model=e2e-mlt.h5

cnzeki/E2E-MLT