This repository provides a modular implementation of SSD object detection in PyTorch 1.0/0.4, supporting MobileNetV1, MobileNetV2, and VGG-based SSD/SSD-Lite architectures. It includes out-of-box support for retraining on the Google Open Images dataset, ONNX export, and Caffe2 conversion for optimized inference. The project follows a design goal of modularity and extensibility.