face.evoLVe

Why we included this project

Teams building identity verification, access control, or any product that matches faces against a known set of people will find a complete toolkit here rather than a single model file. face.evoLVe bundles face detection, landmark localization, affine alignment, and feature extraction, along with training code for backbones like ResNet, IR-SE, and MobileNet and the standard margin-based losses such as ArcFace, CosFace, and SphereFace. That means you can move from aligned training data to a deployed recognition model without stitching together several incompatible repositories. The authors also release pre-trained models and the exact preprocessing and distributed multi-GPU training recipes they used, so you can reproduce published results or fine-tune on your own dataset. Having both PyTorch and PaddlePaddle in one codebase helps teams that work with either framework.