国庆出游，无论是拍人、拍景或是其他，“天空”都是关键元素。比如，一张平平无奇的景物图加上落日余晖的天空色调，氛围感就有了。

当然，自然景观的天空还不是最酷炫的。今天给大家介绍一款基于原生视频的AI处理方法，不仅可以一键置换天空背景，还可以打造任意“天空之城”。

比如换成《星际迷航》中的浩瀚星空、宇宙飞船，将自己随手拍的平平无奇vlog秒变为科幻大片，画面毫无违和感。

该方法源自Github上的开源项目SkyAR，它可以自动识别天空，然后将天空从图片中切割出来，再将天空替换成目标天空，从而实现魔法换天。

下面，我们将基于SkyAR和ModelArts的JupyterLab从零开始“换天造物”。只要脑洞够大，利用这项AI技术，就可以创造出无限种玩法。

本案例在CPU和GPU下面均可运行，CPU环境运行预计花费9分钟，GPU环境运行预计花费2分钟。

实验目标

通过本案例的学习：

了解图像分割的基本应用；

了解运动估计的基本应用；

了解图像混合的基本应用。

注意事项

1. 如果您是第一次使用 JupyterLab，请查看《ModelArts JupyterLab使用指导》了解使用方法；
2. 如果您在使用 JupyterLab 过程中碰到报错，请参考《ModelArts JupyterLab常见问题解决办法》尝试解决问题。

实验步骤

1、安装和导入依赖包

import os
import moxing as mox

file_name = 'SkyAR'
if not os.path.exists(file_name):
    mox.file.copy('obs://modelarts-labs-bj4-v2/case_zoo/SkyAR/SkyAR.zip', 'SkyAR.zip')
    os.system('unzip SkyAR.zip')
    os.system('rm SkyAR.zip')
mox.file.copy_parallel('obs://modelarts-labs-bj4-v2/case_zoo/SkyAR/resnet50-19c8e357.pth', '/home/ma-user/.cache/torch/checkpoints/resnet50-19c8e357.pth')

INFO:root:Using MoXing-v1.17.3-43fbf97f
INFO:root:Using OBS-Python-SDK-3.20.7

!pip uninstall opencv-python -y
!pip uninstall opencv-contrib-python -y

Found existing installation: opencv-python 4.1.2.30
Uninstalling opencv-python-4.1.2.30:
  Successfully uninstalled opencv-python-4.1.2.30
WARNING: Skipping opencv-contrib-python as it is not installed.

!pip install opencv-contrib-python==4.5.3.56

Looking in indexes: http://repo.myhuaweicloud.com/repository/pypi/simple
Collecting opencv-contrib-python==4.5.3.56
  Downloading http://repo.myhuaweicloud.com/repository/pypi/packages/3f/ce/36772cc6d9061b423b080e86919fd62cdef0837263f29ba6ff92e07f72d7/opencv_contrib_python-4.5.3.56-cp37-cp37m-manylinux2014_x86_64.whl (56.1 MB)
     |████████████████████████████████| 56.1 MB 166 kB/s eta 0:00:01|█████▋                          | 9.8 MB 9.4 MB/s eta 0:00:05 MB 9.4 MB/s eta 0:00:05███▏                | 26.6 MB 9.4 MB/s eta 0:00:04/s eta 0:00:03��██▍           | 35.8 MB 9.4 MB/s eta 0:00:03�███████████▌       | 42.9 MB 9.4 MB/s eta 0:00:02��██████████████▎   | 49.6 MB 166 kB/s eta 0:00:40
Requirement already satisfied: numpy>=1.14.5 in /home/ma-user/anaconda3/envs/PyTorch-1.4/lib/python3.7/site-packages (from opencv-contrib-python==4.5.3.56) (1.20.3)
Installing collected packages: opencv-contrib-python
Successfully installed opencv-contrib-python-4.5.3.56
WARNING: You are using pip version 20.3.3; however, version 21.1.3 is available.
You should consider upgrading via the '/home/ma-user/anaconda3/envs/PyTorch-1.4/bin/python -m pip install --upgrade pip' command.

cd SkyAR/

/home/ma-user/work/Untitled Folder/SkyAR
import time
import json
import base64
import numpy as np
import matplotlib.pyplot as plt
import cv2
import argparse
from networks import *
from skyboxengine import *
import utils
import torch
from IPython.display import clear_output, Image, display, HTML


%matplotlib inline

# 如果存在GPU则在GPU上面运行
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
INFO:matplotlib.font_manager:generated new fontManager

2、预览一下原视频

video_name = "test_videos/sky.mp4"


def arrayShow(img):
    img = cv2.resize(img, (0, 0), fx=0.25, fy=0.25, interpolation=cv2.INTER_NEAREST)
    _,ret = cv2.imencode('.jpg', img)
    return Image(data=ret)

# 打开一个视频流
cap = cv2.VideoCapture(video_name)

frame_id = 0
while True:
    try:
        clear_output(wait=True) # 清除之前的显示
        ret, frame = cap.read() # 读取一帧图片
        if ret:
            frame_id += 1
            if frame_id > 200:
                break
            cv2.putText(frame, str(frame_id), (5, 15), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1)  # 画frame_id
            tmp = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # 转换色彩模式
            img = arrayShow(frame)
            display(img) # 显示图片
            time.sleep(0.05) # 线程睡眠一段时间再处理下一帧图片
        else:
            break
    except KeyboardInterrupt:
        cap.release()
cap.release()

3、预览一下要替换的天空图片

img= cv2.imread('skybox/sky.jpg')
img2 = img[:,:,::-1]
plt.imshow(img2)

<matplotlib.image.AxesImage at 0x7fbea986c590>

4、自定义训练参数

可以根据自己的需要, 修改下面的参数

skybox_center_crop: 天空体中心偏移

auto_light_matching: 自动亮度匹配

relighting_factor: 补光

recoloring_factor: 重新着色

halo_effect: 光环效应

parameter = {
  "net_G": "coord_resnet50",
  "ckptdir": "./checkpoints_G_coord_resnet50",

  "input_mode": "video",
  "datadir": "./test_videos/sky.mp4",
  "skybox": "sky.jpg",

  "in_size_w": 384,
  "in_size_h": 384,
  "out_size_w": 845,
  "out_size_h": 480,

  "skybox_center_crop": 0.5,
  "auto_light_matching": False,
  "relighting_factor": 0.8,
  "recoloring_factor": 0.5,
  "halo_effect": True,

  "output_dir": "./jpg_output",
  "save_jpgs": False
}

str_json = json.dumps(parameter)

class Struct:
    def __init__(self, **entries):
        self.__dict__.update(entries)

        
def parse_config():
    data = json.loads(str_json)
    args = Struct(**data)

    return args


args = parse_config()

class SkyFilter():

    def __init__(self, args):

        self.ckptdir = args.ckptdir
        self.datadir = args.datadir
        self.input_mode = args.input_mode

        self.in_size_w, self.in_size_h = args.in_size_w, args.in_size_h
        self.out_size_w, self.out_size_h = args.out_size_w, args.out_size_h

        self.skyboxengine = SkyBox(args)

        self.net_G = define_G(input_nc=3, output_nc=1, ngf=64, netG=args.net_G).to(device)
        self.load_model()

        self.video_writer = cv2.VideoWriter('out.avi',
                                            cv2.VideoWriter_fourcc(*'MJPG'),
                                            20.0,
                                            (args.out_size_w, args.out_size_h))
        self.video_writer_cat = cv2.VideoWriter('compare.avi',
                                                cv2.VideoWriter_fourcc(*'MJPG'),
                                                20.0,
                                                (2*args.out_size_w, args.out_size_h))

        if os.path.exists(args.output_dir) is False:
            os.mkdir(args.output_dir)

        self.output_img_list = []

        self.save_jpgs = args.save_jpgs


    def load_model(self):
        # 加载预训练的天空抠图模型
        print('loading the best checkpoint...')
        checkpoint = torch.load(os.path.join(self.ckptdir, 'best_ckpt.pt'),
                                map_location=device)
        self.net_G.load_state_dict(checkpoint['model_G_state_dict'])
        self.net_G.to(device)
        self.net_G.eval()


    def write_video(self, img_HD, syneth):

        frame = np.array(255.0 * syneth[:, :, ::-1], dtype=np.uint8)
        self.video_writer.write(frame)

        frame_cat = np.concatenate([img_HD, syneth], axis=1)
        frame_cat = np.array(255.0 * frame_cat[:, :, ::-1], dtype=np.uint8)
        self.video_writer_cat.write(frame_cat)

        # 定义结果缓冲区
        self.output_img_list.append(frame_cat)


    def synthesize(self, img_HD, img_HD_prev):

        h, w, c = img_HD.shape

        img = cv2.resize(img_HD, (self.in_size_w, self.in_size_h))

        img = np.array(img, dtype=np.float32)
        img = torch.tensor(img).permute([2, 0, 1]).unsqueeze(0)

        with torch.no_grad():
            G_pred = self.net_G(img.to(device))
            G_pred = torch.nn.functional.interpolate(G_pred,
                                                     (h, w),
                                                     mode='bicubic',
                                                     align_corners=False)
            G_pred = G_pred[0, :].permute([1, 2, 0])
            G_pred = torch.cat([G_pred, G_pred, G_pred], dim=-1)
            G_pred = np.array(G_pred.detach().cpu())
            G_pred = np.clip(G_pred, a_max=1.0, a_min=0.0)

        skymask = self.skyboxengine.skymask_refinement(G_pred, img_HD)

        syneth = self.skyboxengine.skyblend(img_HD, img_HD_prev, skymask)

        return syneth, G_pred, skymask


    def cvtcolor_and_resize(self, img_HD):

        img_HD = cv2.cvtColor(img_HD, cv2.COLOR_BGR2RGB)
        img_HD = np.array(img_HD / 255., dtype=np.float32)
        img_HD = cv2.resize(img_HD, (self.out_size_w, self.out_size_h))

        return img_HD
        

    def process_video(self):
        # 逐帧处理视频
        cap = cv2.VideoCapture(self.datadir)
        m_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
        img_HD_prev = None

        for idx in range(m_frames):
            ret, frame = cap.read()
            if ret:
                img_HD = self.cvtcolor_and_resize(frame)

                if img_HD_prev is None:
                    img_HD_prev = img_HD

                syneth, G_pred, skymask = self.synthesize(img_HD, img_HD_prev)

                self.write_video(img_HD, syneth)

                img_HD_prev = img_HD

                if (idx + 1) % 50 == 0:
                    print(f'processing video, frame {idx + 1} / {m_frames} ... ')

            else:  # 如果到达最后一帧
                break

5、替换天空

替换后输出的视频为out.avi，前后对比的视频为compare.avi

sf = SkyFilter(args)
sf.process_video()

initialize skybox...
initialize network with normal
loading the best checkpoint...
processing video, frame 50 / 360 ... 
processing video, frame 100 / 360 ... 
no good point matched
processing video, frame 150 / 360 ... 
processing video, frame 200 / 360 ... 
processing video, frame 250 / 360 ... 
processing video, frame 300 / 360 ... 
processing video, frame 350 / 360 ... 

6、对比原视频和替换后的视频

video_name = "compare.avi"


def arrayShow(img):
    _,ret = cv2.imencode('.jpg', img)
    return Image(data=ret)

# 打开一个视频流
cap = cv2.VideoCapture(video_name)

frame_id = 0
while True:
    try:
        clear_output(wait=True) # 清除之前的显示
        ret, frame = cap.read() # 读取一帧图片
        if ret:
            frame_id += 1
            cv2.putText(frame, str(frame_id), (5, 15), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1)  # 画frame_id
            tmp = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # 转换色彩模式
            img = arrayShow(frame)
            display(img) # 显示图片
            time.sleep(0.05) # 线程睡眠一段时间再处理下一帧图片
        else:
            break
    except KeyboardInterrupt:
        cap.release()
cap.release()

如果要生成自己的视频，只要将test_videos中的sky.mp4视频和skybox中的sky.jpg图片替换成自己的视频和图片，然后重新一键运行就可以了。赶快来试一试吧，让你的国庆大片更出彩！

华为云社区祝大家国庆节快乐，度过一个开心的假期！

附录

本案例源自华为云AI Gallery：魔幻黑科技，可换天造物，秒变科幻大片！