FMTT: Fused Multi-Head Transformer with Tensor-Compression for 3D Point Clouds Detection on Edge Devices

Zikun Wei, Tingting Wang, Chenchen Ding, Bohan Wang, Ziyi Guan, Hantao Huang, Hao Yu

Abstract: The real-time detection of 3D objects represents a grand challenge on edge devices. Existing 3D point clouds models are over-parameterized with heavy computation load. This paper proposes a highly compact model for 3D point clouds detection using tensor-compression. Compared to conventional methods, we propose a fused multi-head transformer tensor-compression (FMTT) to achieve both compact size yet with high accuracy. The FMTT leverages different ranks to extract both high and low-level features and then fuses them together to improve the accuracy. Experiments on the KITTI dataset show that the proposed FMTT can achieve 6.04× smaller than the uncompressed model from 55.09MB to 9.12MB such that the compressed model can be implemented on edge devices. It also achieves 2.62% improved accuracy in easy mode and 0.28% improved accuracy in hard mode.