FaceX-Zoo is an open-source library on PyTorch for recognizing faces in images. The library provides a module for training models with different configurations of error functionality and basic architecture. In addition, FaceX-Zoo has a standardized module for evaluating trained models on popular datasets for face recognition tasks. The developers published an SDK for directly applying trained neural networks.
Available model configurations
The backbone network extracts features from facial images. FaceX-Zoo allows you to select a core network from a set of architectures:
- MobileFaceNet: a neural network adapted for implementation on mobile devices;
- ResNet: a set of standard architectures for general computer vision tasks;
- SE-ResNet: ResNet, supplemented with SE blocks, which recalibrates features per channel;
- HRNet: Network for Learning High Definition Presentations
Error functionality
FaceX-Zoo contains a set of error functionals that can be used to train models:
- AM-Softmax;
- ArcFace;
- AdaCos;
- AdaM-Softmax;
- CircleLoss;
- CurricularFace;
- MV-Softmax;
- NPCFace;
- Testing Models.
The library also makes it possible to test models on the most popular face recognition datasets: LFW, CPLFW, CPLFW, AgeDB30, RFW, MegaFace, and MegaFace-Mask.