What's that fruit?
summary
Proof of concept shows using Azure Cognitive Services to detect produce at self‑serve checkouts: train a Custom Vision model on about 307 images, get ~91%/88% single-image confidences and up to 99% on edge, exportable as a frozen graph to export to TensorFlow and packaged into a Docker container. Deployment uses IoT Edge to push the container to a Raspberry Pi webcam; the model runs at ~1–2 FPS on Pi, correctly classifies multiple items and enables cloud training plus offline inference for loss‑prevention and accessibility.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 1:15Whats that fruit
- 1:33Who am I
- 2:00Disclaimer
- 3:01Story
- 3:50Carrots
- 4:22Avocados
- 5:26AI
- 6:14Image Classification
- 7:05Azure Cognitive Services
- 7:50Computer Vision
- 10:24Custom Vision
- 11:47Data Preparation
- 12:18Custom Vision AI
- 14:38Adding Images
- 21:12Running on the Edge
- 23:39IRT Edge
- 25:47Dockerfile
- 27:19Docker build
- 30:48Demo
- 36:39Discussion
- 37:55Object Detection
- 39:19Annotation
- 45:36Prediction
- 49:02Recap
related talks
14mComputer vision on edge devices
42mProduct Recognition in Retail/E-Commerce with Azure AI
42mProduct Recognition in Retail/E-Commerce with Azure AI
45mFrom Data to AI on The Edge
58mPlaying the game with Azure Cognitive Services
43mIntersection of Awesome: Building AI enabled IoT solutions powered by the cloud
31mThe Rise of No-Code AI Vision at the Edge
59mAI in the Battle Against Fakes