In this issue, we recommend a one-stop face recognition research platform from JD.com, FaceX-Zoo.
FaceX-Zoo is a PyTorch project for face recognition, including a series of industry-leading solutions such as full-stack face recognition training algorithms, standardized automated test benchmarks, end-to-end face awareness deployment SDK, etc., which can evaluate models in most popular benchmarks with simple configuration, FaceX-Zoo adopts a highly modular design with good scalability and ease of use.
Functional modules
Environmental requirements
python >= 3.7.1
pytorch >= 1.1.0
torchvision >= 0.3.0Identify the results
FaceX-Zoo provides popular backbone networks such as MobileFaceNet, ResNet50-ir, ResNet, TF-NAS-A, HRNet, and EfficientNet to facilitate facial feature extraction. If that’s not enough, just modify the configuration file, add the schema definition file, and you can easily customize any other choice with PyTorch’s support.
Recognition accuracy of different backbone networks on multiple face datasets:
Recognition accuracy of different supervised learning heads on multiple face datasets:
In addition, in order to facilitate the addition of face images wearing masks in the training set, FaceX-Zoo provides the FMA-3D tool to put on masks for the people in the photos:
You can read more on your own.
