This paper addresses the task of semantic scene completion, which involves generating a complete 3D voxel representation of volumetric occupancy and semantic labels from a single-view depth map. The authors observe that scene completion and semantic labeling are closely intertwined, so they introduce SSCNet, an end-to-end 3D convolutional network that jointly predicts occupancy and semantic labels for all voxels in the camera view frustum. To train the network, they create SUNCG, a large-scale synthetic 3D scene dataset with dense volumetric annotations. Experiments show that the joint model outperforms methods addressing each task separately and alternative approaches on the semantic scene completion task. The key contributions include the design of a dilation-based 3D context module for efficient context learning and the construction of SUNCG for comprehensive training data.This paper addresses the task of semantic scene completion, which involves generating a complete 3D voxel representation of volumetric occupancy and semantic labels from a single-view depth map. The authors observe that scene completion and semantic labeling are closely intertwined, so they introduce SSCNet, an end-to-end 3D convolutional network that jointly predicts occupancy and semantic labels for all voxels in the camera view frustum. To train the network, they create SUNCG, a large-scale synthetic 3D scene dataset with dense volumetric annotations. Experiments show that the joint model outperforms methods addressing each task separately and alternative approaches on the semantic scene completion task. The key contributions include the design of a dilation-based 3D context module for efficient context learning and the construction of SUNCG for comprehensive training data.