Last summer, we kicked off the first-ever Xilinx Adaptive Computing Challenge, where we challenged developers and startups to create exciting new applications using Vitis or Vitis AI on select Xilinx hardware platforms. We announced the winners for both the developer and startup contests earlier this year and we're extremely proud of the innovative projects and solutions that were built. We are excited to announce that you can now find the Deepfield-SR solution by BLUEDOT, the first place startup winner, on the new Xilinx App Store!
DeepField-SR: AI-based Super-Resolution
DeepField-SR is a fixed functional hardware accelerator leveraging Xilinx Alveo Cards and AWS F1 to offer the highest computational efficiency for Video Super-Resolution. Based on a proprietary neural network trained with real-world video data from the internet and fusing spatio-temporal information in multiple frames, it produces superior high-resolution video quality.
DeepField-SR is containerized and can be easily run on both AWS F1 and on-premise with Xilinx Alveo U200/U50 Accelerator Cards. The application provides the necessary test scripts and input video files. The upscaled output video files will be generated for you to inspect the video quality and the performance number will be reported in the terminal. This application is also designed in scalable architecture that supports multiple FPGA cards. It can flexibly respond to various resolution upscale requests and its runtime performance on a single Alveo U50 is 11 ~ 14fps to upscale video up to 4K resolution. The API is integrated within an FFmpeg workflow, meaning that simple command enables DeepField-SR acceleration and upscaling user input video.