Core Concepts
This research proposes a novel method for reducing the computational cost of Bird's-Eye-View (BEV) perception models in autonomous driving by selectively pruning redundant sensor data from cameras and LiDAR, achieving comparable performance to state-of-the-art methods while significantly improving efficiency.
Li, Y., Li, Y., Yang, X., Yu, M., Huang, Z., Wu, X., & Yeo, C. K. (2024). Learning Content-Aware Multi-Modal Joint Input Pruning via Bird's-Eye-View Representation. arXiv preprint arXiv:2410.07268.
This paper addresses the computational bottleneck of Bird's-Eye-View (BEV) perception models in autonomous driving by introducing a novel content-aware multi-modal joint input pruning technique. The research aims to reduce the computational overhead of processing sensor data without significantly compromising the accuracy of downstream perception tasks like 3D object detection and map segmentation.