Self-Organizing and Parallel-Process Driven Fast Generation of Adversarial Examples for 3D Point Clouds
TimeWednesday, July 13th6pm - 7pm PDT
LocationLevel 2 Lobby
Event Type
Networking Reception
Work-in-Progress Poster
DescriptionWe propose a new data-driven fast adversarial sample generation method for 3D point cloud networks. Input the original point cloud into the fast-global-feature-extraction network to obtain the global feature of the point cloud. The fast-global-feature-extraction network uses the self-organizing map to guide the point cloud for group coding, which speeds up the coding process and reduces resource usage. Then input the global features into the fast-point-cloud-recovery network to obtain a point cloud with adversarial properties. The fast-point-cloud-recovery network has dual parallel branches, which can quickly obtain highly effective adversarial samples under the premise of low resource occupation.