Face recognition using computer vision technology has now matured, there are many outstanding items on Gitee face recognition, but the outstanding developers on Gitee reverse thinking, the use of computer vision to achieve real-time video portrait of the Web side erased, so that you become invisible in the eyes of the camera.
Project name: Web-rmbody
Project author: Zhang Zijian
Open-source license: the Apache-2.0
Introduce the principle (from author Readme)
Google has developed a face recognition model BodyPix, real time video in all portraits marked. As the "three-body" in the "ideological stamp," If we give the results of the model of a "+" number, that is to everyone identified, but if given a "-" No, you can put these people from the background in erased.
In this project, the algorithm according to the model results, there is no real time portrait of the write buffer portion, and the use of alternate data buffer area having the same position in place of the portrait. Thus requiring the video scene must be stationary (stationary camera recommended), and all context must not be covered at the beginning of the portrait.
Specific experience the effect depends on device performance, I tested on my phone Redmi K20 Pro (Snapdragon 855, 6GB RAM), sense of self, can more smoothly use (there should be 10 FPS); but in the iPad mini 2 (Apple A7, the test 1GB RAM) found it difficult to use normal (2-3 FPS). Runs on most PC should be no problem.
The project is still in the experimental stage, and the effect may be poor.
Online Demo Address: https://fsyz.online/demo/Web-rmbody/
Demonstration effect:
The project is still at an experimental stage, or to welcome students interested bigwigs who has studied computer vision project go home, building blocks for this project: https://gitee.com/futrime/Web-rmbody
