Living on the Cloud's Edge
summary
Built a local video pipeline that counts people while keeping raw faces off-network, running models on small boards and USB accelerators. Uses Coral Edge TPU and prototyping boards like NVIDIA Jetson or Raspberry Pi plus a USB accelerator; CPU-only drops to ~2 FPS, TPU/GPU runs jump to tens–hundreds FPS (examples: 120 FPS for a lightweight face model, ~70–80 FPS for person detection, GPU spikes to ~770 in demo). Prototype costs ~USD 200 plus $70 for a Coral USB stick; demos work but have occasional reliability and model-training wrinkles.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:38Building something...
- 5:20Internet of Things
- 13:48What is Fog Computing?
- 15:28Edge Computing Devices
- 15:55Edge AI Computing
- 16:31AI Accelerators
- 18:01Google Coral Dev Board
- 23:42Detection vs. Tracking
related talks
29mLife on the Edge
49mEnable IoT with Edge Computing and Machine Learning
48m50 Billion Connected Things are Coming
39mExploring the Trend Toward the Edge
25mIoT Compute Unleashed: Transparent Offload From ESP32 To Edge Clusters
41mCloud Computing: The Edge talks - long road into the Cloud
17mAI workflow for large scale deployment of far-edge ML devices
36mWebAssembly-based Edge AI Sensing Pipeline Orches...