Extracting Information from Rich Video Streams: An Agile Software/Hardware Approach
Abstract
Over the course of 4.5 years, this project has worked to enhance the state of the art in image processing and, by extension, machine learning algorithms and hardware. The projects' four subgroups each focused on a specific set of tasks. 1) The applications group developed a suite of video-processing applications aimed at running with high efficiency on a coarse-grained reconfigurable architecture (CGRA)-based SoC. 2) The tools group produced compilers and generators capable of automatically moving these Halide-coded applications down to efficient hardware, such as the aforementioned CGRA. 3) The hardware group designed and taped out multiple generations of a CGRA-based SoC capable of efficiently running these applications. And 4) a test and validation group made sure that collateral was both correct and robust. This report summarizes progress made in each of these areas.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 2023
- Accession Number
- AD1200191
Entities
People
- Mark Horowitz
- Stephen Richardson
Organizations
- Stanford University