Improvement of 3D Semantic Completion from a Single Depth Image

Project aims at completing the 3D semantic scene from the depth images.

• We introduces a post-processing module to improve the Average Precision (AP) and Intersection-over-union (IoU).
• Because the cost of 3D-CNN is tremendous, we redesign the network to make it fit in the GPU memory(Nvidia).
• The aggregated end-to-end network improves the AP by 6.1% and IoU by 2.3%. [Notes]

• Softwares: Ubuntu 14.04 TLS, Caffe, Matlab 2016a or above
• Hardwares: NVIDIA GeForce GTX Titan X (12GB), RAM 32GB

[1] S. Song, F. Yu, A. Zeng, A. X. Chang, M. Savva, and T. Funkhouser, "Semantic scene completion from a single depth image," in Proceedings of 30th IEEE Conference on Computer Vision and Pattern Recognition, 2017.