Python3.7 老照片修复算法

https://github.com 可更换为 https://github.com.cnpmjs.org 或 https://hub.fastgit.org 镜像

github加速参考 https://mp.weixin.qq.com/s/L1Xz9lXoibCsiJpaShohpQ

安装Python3.7

下载地址：https://www.python.org/downloads 一、修复照片划痕，清晰度 1.克隆项目地址 ```shell script git clone https://github.com/microsoft/Bringing-Old-Photos-Back-to-Life

2.克隆BatchNorm PyTorch库到脸部和全局

cd Face_Enhancement/models/networks/ git clone https://github.com/vacancy/Synchronized-BatchNorm-PyTorch cp -rf Synchronized-BatchNorm-PyTorch/sync_batchnorm . cd ../../../

cd Global/detection_models git clone https://github.com/vacancy/Synchronized-BatchNorm-PyTorch cp -rf Synchronized-BatchNorm-PyTorch/sync_batchnorm . cd ../../

3.下载landmark detection预训练模型

cd Face_Detection/ wget http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2 bzip2 -d shape_predictor_68_face_landmarks.dat.bz2 cd ../

4.下载预训练库

cd Face_Enhancement/ wget https://github.com/microsoft/Bringing-Old-Photos-Back-to-Life/releases/download/v1.0/face_checkpoints.zip unzip face_checkpoints.zip cd ../ cd Global/ wget https://github.com/microsoft/Bringing-Old-Photos-Back-to-Life/releases/download/v1.0/global_checkpoints.zip unzip global_checkpoints.zip cd ../

5.安装依赖包

pip install -r requirements.txt

> 备注
>
> dlib依赖包失败可使用以下依赖包[dlib-19.17.99-cp37-cp37m-win_amd64.whl](dlib-19.17.99-cp37-cp37m-win_amd64.whl)
> 
> 安装本地依赖包
>
> pip install dlib-19.17.99-cp37-cp37m-win_amd64.whl
>
> 安装指定依赖包
>
> pip install dlib -i https://pypi.doubanio.com/simple
> 
> 
7.测试

// 没有划痕的图像： python run.py –input_folder ./input –output_folder ./output –GPU -1 // 有划痕图像： python run.py –input_folder ./input –output_folder ./output –GPU -1 –with_scratch // 有划痕的高分辨率图像： python run.py –input_folder ./input –output_folder ./output –GPU -1 –with_scratch –HR

二、照片上色
1.克隆项目地址

git clone https://github.com/jantic/DeOldify

2.下载训练模型

下载Artistic [https://data.deepai.org/deoldify/ColorizeArtistic_gen.pth](https://data.deepai.org/deoldify/ColorizeArtistic_gen.pth)

下载Stable(须科学上网) [https://www.dropbox.com/s/usf7uifrctqw9rl/ColorizeStable_gen.pth?dl=0](https://www.dropbox.com/s/usf7uifrctqw9rl/ColorizeStable_gen.pth?dl=0)

下载Video [https://data.deepai.org/deoldify/ColorizeVideo_gen.pth](https://data.deepai.org/deoldify/ColorizeVideo_gen.pth)

    | deoldify
      | models
        | ColorizeVideo_gen.pth (艺术)
        | ColorizeStable_gen.pth (正式)
        | ColorizeArtistic_gen (视频)

3.安装依赖包

pip install -r requirements.txt

> 备注：Bottleneck依赖包下载失败可使用附件地址下载
>
4.启动jupyterlab服务

jupyter lab

5.创建文件run.ipynb

#NOTE: This must be the first call in order to work properly! from deoldify import device from deoldify.device_id import DeviceId #choices: CPU, GPU0…GPU7 device.set(device=DeviceId.CPU)

import torch

if not torch.cuda.is_available(): print(‘GPU not available.’)

import fastai from deoldify.visualize import * import warnings warnings.filterwarnings(“ignore”, category=UserWarning, message=”.?Your .? set is empty.*?”)

colorizer = get_image_colorizer(artistic=True)

colorizer.plot_transformed_image(“test_images/1.png”, render_factor=10, compare=True)

```

附件下载：

源码百度网盘下载：

https://pan.baidu.com/s/1RBFoC2t-48zJ9jARijZ7bA?pwd=22wr

dlib依赖包下载：

https://file.alonesky.com/file/python/dlib-19.17.99-cp37-cp37m-win_amd64.whl

Bottleneck依赖包下载：

https://file.alonesky.com/file/python/Bottleneck-1.3.2-cp37-cp37m-win_amd64.whl

参考网址：

https://mp.weixin.qq.com/s/sTaj2ZtoWoIT0ViR3onEpw

https://github.com.cnpmjs.org/microsoft/Bringing-Old-Photos-Back-to-Life