SECOND (Sparsely Invoked Convolution-based Detector) is a 3D object detection framework that uses sparse convolution and VoxelNet architecture to detect objects from point cloud data. It supports training on KITTI and NuScenes datasets, implements PointPillars and sparse convolution optimization, and provides training pipelines with fp16 and multi-GPU support. The project is deprecated in favor of OpenPCDet and mmdetection3d.