Empowering Knowledge Distillation via Open Set Recognition for Robust 3D Point Cloud Classification

Real-world scenarios pose several challenges to deep learning based computer\nvision techniques despite their tremendous success in research. Deeper models\nprovide better performance, but are challenging to deploy and knowledge\ndistillation allows us to train smaller models with minimal loss in\nperformance. The model also has to deal with open set samples from classes\noutside the ones it was trained on and should be able to identify them as\nunknown samples while classifying the known ones correctly. Finally, most\nexisting image recognition research focuses only on using two-dimensional\nsnapshots of the real world three-dimensional objects. In this work, we aim to\nbridge these three research fields, which have been developed independently\nuntil now, despite being deeply interrelated. We propose a joint Knowledge\nDistillation and Open Set recognition training methodology for\nthree-dimensional object recognition. We demonstrate the effectiveness of the\nproposed method via various experiments on how it allows us to obtain a much\nsmaller model, which takes a minimal hit in performance while being capable of\nopen set recognition for 3D point cloud data.\n