What is Deepfakes, in simple terms, it is to replace the face of a person in a video with the face of another person
Some time ago, some people used this technology to synthesize love action movies, but to be honest, the technology is innocent. You can't stop driving because the car will crash and kill people. The technology is developing, and I believe this technology will be in the near future. Can give off its true light.
Furthermore, this technology has extremely high requirements on the quality and even the quantity of hardware, video, and face-changing pictures. This is a strong reason to persuade them to quit. Let’s talk about hardware first, try to 1060 online, and have at least 8g of memory (of course, I said this). is the standard configuration, if you ask if you can run below this configuration? Of course you can, but it’s really hard to say what the effect is), under this configuration, the picture training time is at least 6 hours.
As for the video requirements, in fact, the video is also split into pictures in this technology, so it is basically the same as the picture requirements of the face-changing person. A person's face, if there are other people's faces, it needs to be removed (for example, after splitting into pictures, delete the relevant pictures)
The most difficult thing to do is actually the pictures of face-changing people. We need to have 1000-5000 pictures of face-changing people as much as possible, and there can only be one face of the face-changing person in the picture. If we want the face-changing effect to be good, we will also We need photos of the person's face under different lighting conditions. Of course, we can also use a video of a face-changing person for video processing and split it into many photos.
Let’s talk about the function of Fake app, the main hero of video face changing, and then let’s talk about the technical steps:
Fake app is an image processing software based on deep learning , not a software that directly processes video as we imagine.
So this face changing process can be roughly summarized as:
Use video processing software for the video that needs to be changed, and divide it into many, many pictures by frame, then collect the pictures of our face-changing people, use the Fake app to start training, and generate a new bunch of pictures. At this time, these pictures are actually It is the split picture of the new video, and then use the video processing software to synthesize the picture into a new video.
Please watch the video for the actual operation process:
Deepfakes1.1 teaching video
https://tieba.baidu.com/p/5541202771
Fakeapp official website
fakeapp.org