FractalCloud: A Fractal-Inspired Architecture for Efficient Large-Scale Point Cloud Processing
This program is tentative and subject to change.
Three-dimensional (3D) point clouds are increasingly used in applications such as autonomous driving, robotics, and virtual reality (VR). Point-based neural networks (PNNs) have demonstrated strong performance in point cloud analysis, originally targeting small-scale inputs. However, as PNNs evolve to process large-scale point clouds with hundreds of thousands of points, global computation and memory access in point cloud processing introduce substantial overhead, causing quadratically increased latency. Existing accelerators, primarily optimized for small-scale workloads, overlook this challenge and scale poorly due to inefficient partitioning and non-parallel architectures. To address these issues, we propose FractalCloud, a fractal-inspired hardware architecture for efficient large-scale 3D point cloud processing. FractalCloud introduces two key optimizations: (1) a fractal method for shape-aware and hardware-friendly partitioning, and (2) block-parallel point operations that decompose and parallelize all point operations. A dedicated hardware design with on-chip fractal and flexible parallelism further enables fully parallel processing within limited memory resources. Implemented in 28 nm technology as a chip layout with a core area of 1.5 $mm^2$, FractalCloud achieves 21.7× speedup and 27× energy reduction over state-of-the-art accelerators while maintaining network accuracy, demonstrating its scalability and efficiency for PNN inference.
This program is tentative and subject to change.
Mon 2 FebDisplayed time zone: Hobart change
15:50 - 17:10 | |||
15:50 20mTalk | GRTX: Efficient Ray Tracing for 3D Gaussian-Based Rendering Main Conference Junseo Lee Seoul National University, Sangyun Jeon Seoul National University, Jungi Lee Seoul National University, Junyong Park Seoul National University, Jaewoong Sim Seoul National University | ||
16:10 20mTalk | Splatonic: Architecture Support for 3D Gaussian Splatting SLAM via Sparse Processing Main Conference Xiaotong Huang Shanghai Jiao Tong University, He Zhu Shanghai Jiao Tong University, Tianrui Ma Institute of Computing Technology, Chinese Academy of Sciences, Yuxiang Xiong Shanghai Jiao Tong University, Fangxin Liu Shanghai Jiao Tong University, Zhezhi He Shanghai Jiao Tong University, Yiming Gan Institute of Computing Technology, Chinese Academy of Sciences, Zihan Liu Shanghai Jiao Tong University, Jingwen Leng Shanghai Jiao Tong University, Yu Feng Shanghai Jiao Tong University, Minyi Guo Shanghai Jiao Tong University | ||
16:30 20mTalk | FractalCloud: A Fractal-Inspired Architecture for Efficient Large-Scale Point Cloud Processing Main Conference Yuzhe Fu Duke University, Changchun Zhou Duke University, Hancheng Ye Duke University, Bowen Duan Duke University, Qiyu Huang Yale University, Chiyue Wei Duke University, Cong Guo Duke University, Hai "Helen" Li Duke University, Yiran Chen Duke University | ||
16:50 20mTalk | ORANGE: Exploring \underline{O}ckham's \underline{R}azor for Neural Rendering by \underline{A}ccelerating 3DGS on \underline{N}PUs with \underline{GE}MM-Friendly Blending and Balanced Workloads Main Conference Haomin Li Shanghai Jiao Tong University, Yue Liang Shanghai Jiao Tong University, Fangxin Liu Shanghai Jiao Tong University, Bowen Zhu Shanghai Jiao Tong University, Zongwu Wang Shanghai Jiao Tong University, Yu Feng Shanghai Jiao Tong University, Liqiang Lu Zhejiang University, Li Jiang Shanghai Jiaotong University, Haibing Guan Shanghai Jiao Tong University | ||