Using Augmented Face Images to Improve Facial Recognition Tasks
Authors: Shuo Cheng, Guoxian Song, Wan-Chun Ma, Chao Wang, Linjie Luo
Published: 2022-05-13 20:12:12+00:00
AI Summary
This paper proposes a framework to improve facial recognition by augmenting training data with GAN-generated images. The approach focuses on adding underrepresented attributes like eyewear or beards while preserving facial expressions, enhancing model robustness.
Abstract
We present a framework that uses GAN-augmented images to complement certain specific attributes, usually underrepresented, for machine learning model training. This allows us to improve inference quality over those attributes for the facial recognition tasks.
Key findings
The results show that training with the augmented data improves the performance of facial recognition models, particularly in handling underrepresented attributes. Filtering out low-quality generated images further enhances the model's performance.
Approach
The authors use a pre-trained StyleGAN2 and InterFaceGAN to generate augmented face images by manipulating latent codes representing specific attributes. A quality assessment based on landmark detection filters out low-quality augmentations before training a CNN-based model for facial blendshape activation prediction.
Datasets
UNKNOWN
Model(s)
StyleGAN2, InterFaceGAN, CNN with fully connected layers
Author countries
USA