Nvidia unveils low-power chip for sub-millisecond face detection

Nvidia researchers have developed an always-on vision system capable of detecting human faces in under a millisecond. This system, integrated into a chip, consumes less than 5 milliwatts while operating at 60 frames per second, a significant reduction from typical systems requiring around 10 watts. The technology utilizes a "race to sleep" approach, where a deep learning accelerator performs detection rapidly and then the system enters a low-power state, making it suitable for applications like autonomous vehicles and power-saving consumer electronics. AI

排序理由 Nvidia researchers presented a new efficient vision-processing system at a conference, detailing its technical specifications and potential applications.

在 IEEE Spectrum — AI 阅读 →

• Alpha-Vision
• Ben Keller
• IEEE International Solid State Circuits Conference
• Nvidia
• San Francisco
• SRAM

• 基础设施
• 论文
• 产品

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →