Live Streaming Object Detection by Leveraging GPU with AVA Platform on EWAOL
summary
End-to-end pipeline for live-streaming object detection runs containers built on AWS EC2 g5g (ARM + AWS Graviton2) and pushed to Docker Hub, then deployed to an AVA edge platform paired with NVIDIA GeForce RTX 3080; an RTSP server on Raspberry Pi feeds live camera to the edge object-detector container (NGC images with CUDA). Detection accuracy matches cloud vs edge per-frame, measured ~24 FPS on cloud vs ~15 FPS on edge; issues: AVA lacks direct camera support and container images are very large.
AI-written summary · about
chapters (9)
- 0:00Live Streaming Object Detection by Leveraging GPU with AVA Platform on EWAOL
- 1:15Cloud Edge Software Architecture
- 3:00Architecture for Edge Side
- 3:51How to use NVIDIA GPU to run AI model under Docker
- 4:19Hardware setup of Edge Side
- 4:44Demo of Perform Object Detection with Video File on Cloud
- 8:26Cloud and Edge Object detection results
- 11:12Demo – Perform Object Detection with Live Camera on Edge
- 12:20Demo – Live Camera Object Detection result
related talks
53mVideo Object Detection at Scale
53mVideo Analytics for mere mortals with NVIDIA DeepStream SDK
15mReal-Time Object Detection with Pravega and Flink
09mLightning Talk: Streaming and Processing Edge Vision Data in Real Time
06mLightning Talk: Deploying AI Solutions on the Edge with Containers
01h 56mEdge AI Systems
45mExtracting real-time data at the edge with SageMaker & NVIDIA GPUs
36mVideo analytics at scale: Deep Learning, Computer Vision, ML & Spark