1 paper · 1 filter
Atrin Arya, Hanieh Naderi, Shohreh Kasaei
Recent research has revealed that the security of deep neural networks that directly process 3D point clouds to classify objects can be threatened by adversarial samples. Although…