GFPGAN:用于真实世界面部修复的实用算法

2024-07-25

计算机视觉

字数统计: 1.1k | 阅读时长≈ 5 分钟

一、GFPGAN

（1）介绍

官网地址：TencentARC/GFPGAN: GFPGAN aims at developing Practical Algorithms for Real-world Face Restoration. (github.com)

这个GFPGAN是一种图像修复的算法，可以把破旧模糊的图片还原成清晰可见的照片。

一点也不夸张，我试了一下，把一个64x64的照片变成256x256分辨率的清晰照片，非常高清。

这个算法是基于这个叫Real-ESRGAN的项目，所以都是一步步迭代上来的，挺了不起的！感兴趣的朋友可以去复现一下。这里面还有需要注意的两个项目包，前面的real-rsrgan和gfpgan都是基于下面两个基础上完成的：

BasicSR: An open-source image and video restoration toolbox
facexlib: A collection that provides useful face-relation functions

basicsr负责的是图像和视频修复，这个是最厉害的工具，而facexlib是提供面部关系一个函数库，提高修复效果。
论文：GFP-GAN: Towards Real-World Blind Face Restoration with Generative Facial Prior

（2）安装环境

1）主要是python≥3.7、pytorch≥1.7、一块GPU显卡

2）下载命令：

1 2	git clone https://github.com/TencentARC/GFPGAN.git cd GFPGAN

# Install basicsr - https://github.com/xinntao/BasicSR
# We use BasicSR for both training and inference
pip install basicsr

# Install facexlib - https://github.com/xinntao/facexlib
# We use face detection and face restoration helper in the facexlib package
pip install facexlib

pip install -r requirements.txt
python setup.py develop

# If you want to enhance the background (non-face) regions with Real-ESRGAN,
# you also need to install the realesrgan package
pip install realesrgan

3）下载模型GFPGANv1.3（可选项，也可以是v1.2和v1）：

1	wget https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth -P experiments/pretrained_models

4）运行代码，run：

1	python inference_gfpgan.py -i inputs/whole_imgs -o results -v 1.3 -s 2

命令的一些参数：

Usage: python inference_gfpgan.py -i inputs/whole_imgs -o results -v 1.3 -s 2 [options]...

  -h                   show this help
  -i input             Input image or folder. Default: inputs/whole_imgs
  -o output            Output folder. Default: results
  -v version           GFPGAN model version. Option: 1 | 1.2 | 1.3. Default: 1.3
  -s upscale           The final upsampling scale of the image. Default: 2
  -bg_upsampler        background upsampler. Default: realesrgan
  -bg_tile             Tile size for background sampler, 0 for no tile during testing. Default: 400
  -suffix              Suffix of the restored faces
  -only_center_face    Only restore the center face
  -aligned             Input are aligned faces
  -ext                 Image extension. Options: auto | jpg | png, auto means using the same extension as inputs. Default: auto

二、项目结构和代码介绍

（1）代码地址

https://github.com/Lucki-ly/GFPGAN_1

（2）项目结构

如上图所示，项目的结构有这么多，解释如下：
1）experiments：存放pretrained_models，就是我们的模型（GFPGANv1.3.pth等）；

2）gfpgan：模型的目录；

3）inputs：输入文件的目录；

4）options：存放一些训练的yaml文件；

5）results：输出文件的目录；

6）tests：存放一些测试的文件，测试功能；

7）inference_gfpgan.py：项目接口的运行主文件，我们靠这个调用所有的功能；

8）requirements.txt：项目要求的安装文件；

9）setup.py：项目的启动文件，运行之前要python一下。

（3）代码调整

我做了一个图像尺寸（分辨率）的控制，这样我们可以根据自己的需求去获得想要的大小。

前面的参数增加：

1	parser.add_argument('-d', '--desired_size', type=int, default=512, help='Desired size for the output images. Default: 256')

restore部分修改：

 # ------------------------ restore ------------------------
for img_path in img_list:
    # read image
    img_name = os.path.basename(img_path)
    print(f'Processing {img_name} ...')
    basename, ext = os.path.splitext(img_name)
    input_img = cv2.imread(img_path, cv2.IMREAD_COLOR)
        
        
    # Print the size of the input image
    height, width, _ = input_img.shape
    print(f'Input image size: {width}x{height}')


    # restore faces and background if necessary
    cropped_faces, restored_faces, restored_img = restorer.enhance(
        input_img,
        has_aligned=args.aligned,
        only_center_face=args.only_center_face,
        paste_back=True,
        weight=args.weight)

    # Print the size of the restored image
    if restored_img is not None:
        height, width, _ = restored_img.shape
        print(f'Restored image size: {width}x{height}')
            
        # Resize the restored image to 256x256
        restored_img = cv2.resize(restored_img, (args.desired_size, args.desired_size), interpolation=cv2.INTER_LINEAR)
            
        # Print the size of the restored image after resizing
        height, width, _ = restored_img.shape
        print(f'Restored image size after resizing: {width}x{height}')
            
    # save faces
    for idx, (cropped_face, restored_face) in enumerate(zip(cropped_faces, restored_faces)):
        cropped_face = cv2.resize(cropped_face, (args.desired_size, args.desired_size), interpolation=cv2.INTER_LINEAR)
        restored_face = cv2.resize(restored_face, (args.desired_size, args.desired_size), interpolation=cv2.INTER_LINEAR)
            
        # save cropped face
        save_crop_path = os.path.join(args.output, 'cropped_faces', f'{basename}_{idx:02d}.jpg')
        imwrite(cropped_face, save_crop_path)
        # save restored face
        if args.suffix is not None:
            save_face_name = f'{basename}_{idx:02d}_{args.suffix}.jpg'
        else:
            save_face_name = f'{basename}_{idx:02d}.jpg'
        save_restore_path = os.path.join(args.output, 'restored_faces', save_face_name)
        imwrite(restored_face, save_restore_path)
        # save comparison image
        cmp_img = np.concatenate((cropped_face, restored_face), axis=1)
        imwrite(cmp_img, os.path.join(args.output, 'cmp', f'{basename}_{idx:02d}.jpg'))