PVNAS: 3D Neural Architecture Search with Point-Voxel Convolution
This work addresses the need for fast and accurate 3D deep learning on resource-limited edge devices, such as in AR/VR and self-driving cars, by introducing a novel hybrid approach that improves efficiency and performance.
The paper tackles the problem of inefficient 3D neural networks on edge devices by proposing a hardware-efficient primitive, Point-Voxel Convolution, and using neural architecture search to optimize networks under resource constraints, achieving state-of-the-art performance with 1.8-23.7x speedup on benchmark datasets and improved deployment in autonomous vehicles.
3D neural networks are widely used in real-world applications (e.g., AR/VR headsets, self-driving cars). They are required to be fast and accurate; however, limited hardware resources on edge devices make these requirements rather challenging. Previous work processes 3D data using either voxel-based or point-based neural networks, but both types of 3D models are not hardware-efficient due to the large memory footprint and random memory access. In this paper, we study 3D deep learning from the efficiency perspective. We first systematically analyze the bottlenecks of previous 3D methods. We then combine the best from point-based and voxel-based models together and propose a novel hardware-efficient 3D primitive, Point-Voxel Convolution (PVConv). We further enhance this primitive with the sparse convolution to make it more effective in processing large (outdoor) scenes. Based on our designed 3D primitive, we introduce 3D Neural Architecture Search (3D-NAS) to explore the best 3D network architecture given a resource constraint. We evaluate our proposed method on six representative benchmark datasets, achieving state-of-the-art performance with 1.8-23.7x measured speedup. Furthermore, our method has been deployed to the autonomous racing vehicle of MIT Driverless, achieving larger detection range, higher accuracy and lower latency.