This repository provides MobileNetV2-YOLOv3-Lite and Nano variants optimized for real-time object detection on mobile devices. It implements depthwise separable convolutions from MobileNetV2 to reduce computational cost (0.5BFlops) while maintaining reasonable accuracy. The project supports face detection with landmark regression and provides C language samples for NCNN deployment, targeting ARM processors on devices like Huawei P40.