Face key point data enhancement method

Abstract

The invention belongs to the technical field of face recognition, and provides a face key point data enhancement method, device, system and model training method, through the face key point data enhancement method through affine transformation and splicing of the local area of ​​the face image, with Increasing the number and diversity of samples is conducive to improving the training effect of the key point detection model. On the basis of labeled data, face key point data can be efficiently generated by using local affine transformation, thereby improving the accuracy and generalization of subsequent model training.

Description

technical field [0001] The present invention relates to the technical field of face recognition, in particular to a face key point data enhancement method based on local affine transformation, a device, a system and a method for training a face key point detection model. Background technique [0002] Training the face key point detection model requires large-scale labeled face data. Most of the obtained face images are natural expressions, such as relatively few images with mouth open and eyes closed. This relatively single data will make training The generalization ability of the latter model is poor. [0003] In addition, the annotation of face key point data is time-consuming and laborious, and annotators need to finely mark dozens or even hundreds of key points for each face image. These two factors greatly limit the acquisition of high-quality data, which in turn affects the performance of deep learning models. Data enhancement is an efficient method to quickly increa...

AI Technical Summary

Problems solved by technology

[0002] Training the face key point detection model requires large-scale labeled face data. Most of the obtained face images are natural expressions, such as relatively few images with mouth open and eyes closed. This relatively single data will make training After the model generalization ability is poor

[0003] In addition, the annotation of face key point data is time-consuming and laborious, and annotators need to carefully mark dozens or even hundreds of key points for each face image

These two factors greatly limit the acquisition of high-quality data, which in turn affects the performance of deep learning models.

Data enhancement is an efficient method to quickly increase the amount of data, but the existing face key point data enhancement methods, such as translation, rotation, zooming, etc., cannot change the contour and facial features of the face, such as eyes open to closed eyes, fat thin face shape

So even though the amount of data has increased, the diversity of the data has not increased

Images