This repository provides a PyTorch implementation of RetinaFace, a single-stage dense face localization model originally published by deepinsight/insightface. The model performs face detection and localization on images, achieving results such as 89.60% on WIDERFace hard val with ResNet50 backbone and 80.99% with the smaller MobileNet0.25 backbone. The project includes training code, evaluation scripts, and TensorRT deployment support for inference.