⚡ 本页包含 AI 生成的分析内容，仅供参考
ALPhA-Vision是一种始终开启的低功耗视觉处理器子系统，专为边缘SoC中的DNN推理任务设计。它通过硬件/软件协同设计支持CNN和ViT推理，实现无需外部存储器的端到端执行，并以787μs延迟和99.3%精度完成人脸检测，功耗低于5mW。
Qijing Huang1, Shalini De Mello1, Brucek Khailany2 Nvidia, Santa Clara, CA, 2Nvidia, Austin, TX, 3Nvidia, Durham, NC, 4Stanford University, Stanford, CA 1 Abstract ALPhA-Vision is an always-on low-power subsystem for DNN-inference-based vision tasks in edge SoCs. Flexible and programmable, the subsystem supports CNN and ViT inference and employs hardware/software co-design to enable fully end-to-end execution with no external memory accesses. Fine-grained power management features to mitigate leakage enable the subsystem to perform face detection with 787μs latency and 99.3% detection accuracy with 4.6mW average power at 60fps. Always-on computer vision (CV) tasks are key workloads in applications such as user presence detection, safety monitoring, and gesture-based control, with always-on vision (AoV) poised to become even more prevalent as embodied artificial intelligence enters the mainstream [1]. In consumer devices, use cases such as face detection for opportunistic
Ben Keller1, Rangharajan Venkatesan1, Steve Dai1, Jason Clemons2, Matthew Fojtik3, Muya Chang1, Thierry Tambe4, Nathaniel Pinckney2, Stephen G. Tell3,