FacePose_pytorch

WIKI2020/FacePose_pytorch is ranked 14,070 among these 8 peers; facebookresearch/PoseDiffusion is ranked 14,039.

FacePose_pytorch The pytorch implement of the head pose estimation(yaw,roll,pitch) and emotion detection with SOTA performance in real time. Easy to deploy, easy to use, and high accuracy. Solve all problems of face detection at one time. 7 pip install mediapipe python newdectect. y (run on cpu) You can replace the pfld algorithm(this GitHub) by yourself. The algorithm has been applied to the following two products: online education for children, which is used to identify whether

children listen carefully; on-site meeting or school classroom, to judge the quality of speech. The head angle, especially the accuracy of facial emotion recognition, has reached the world's top level. However, for some reasons, we only open source prediction code. We will release the ultra-high precision model in future(Including angles and emotion). If you need, please add a github star and leave email, I will send it to you separately. Demo Training There is no need to

model train(Using my model is enough) Introduction 1. Firstly, the Retinaface is used to extract the face frame, and then PFLD is used to identify the key points of the face. Finally, the key points are followed up to estimate the face pose. It is very easy to deploy and use, with high precision and fast speed. We collected our own facial angle conversion data from hundreds of people and fit a simple linear model through the rotation key

points of hundreds of people's faces. Experiments show that the simple mathematical linear point model is more efficient and accurate(You can also use GBDT or other algorithms to regression). At the same time, referring to several papers published in the United States summit in 2020, we developed a highly accurate emotion recognition model. Results show that the methods with 95% on raf-db, 80% on affectnet, and 98% on ferplus. At present, we predict seven kinds of expressions: "surprise

", "fear", "strange", "happiness", "Sadness", "anger" and "neutral". Performance Example Other Project The following is a list of some classic face pose detection items, you can compare the effect by yourself OpenFace:[ Dlib:[ 3DDFA_V2:[ FAQ **Q1. Why implement this while there are several FacePose projects already. ** A1: Because the existing open source project identification error is big. And this is a Key point detection and expression detection integration project. **Q2: What exactly is the difference among this repository and the

others? ** A2: For example, Here are some of the common methods used by other open source projects: 1. Dlib:It is not accurate for face key points recognition, and the error is large when the face is rotated or roll. Virtual 3D model:it is very inaccurate to compare the recognition of key points with a "2D to 3D Virtual fix model",because everyone has a different face. Through the convolution network statistics face key point bitmap, the angle is

also very inaccurate. **Q3: What should I do when I find a bug? ** A3: Check out the update log if