Streamed Video Processing for Cloud-Scale Vision AI Services

Real-time, cloud-scale applications for AI-based computer vision (CV) are growing rapidly, with video data now becoming the primary component of consumer internet traffic. Therefore, building high-performance and cost-effective cloud solutions for video processing workloads is critical. Existing solutions have high interoperable cost among heavyweight frameworks for video encoding/decoding, pre-/post-processing, and inference. As a paradigm shift, we'll dive deep into a streamed video processing prototype by leveraging NVIDIA’s Triton inference server and lightweight NVIDIA CV core components, demonstrating faster deployment and better scalability. Also, Netflix experts will join us to talk about their novel video applications using the optimizations.