GTC 2020: Deep Learning for 3D Vision (Point Clouds)
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Deep Learning for 3D Vision (Point Clouds)
Winston Hsu, National Taiwan University
Learning on 3D point clouds is vital for a broad range of emerging applications such as autonomous driving, robot perception, VR/AR, gaming, and security. Such needs have increased recently due to the prevalence of 3D sensors such as lidar, 3D cameras, and RGB-D depth sensors. Point clouds consist of thousands to millions of points and are complementary to the traditional 2D cameras in the vision (or multimedia) community. 3D learning algorithms on point cloud data are new, and exciting, for numerous core problems such as 3D classification, detection, semantic segmentation, and face recognition. The tutorial covers the 3D sensors, 3D representations, emerging applications, core problems, state-of-the-art learning algorithms (for example, voxel-based and point-based), and future research opportunities. We'll also showcase our leading work in several 3D benchmarks such as ScanNet, KITTI, etc., and efficient neural network training (with data parallelism) by NVIDIA GPU platforms (for example, DGX-1).