Support Silicon Dojo at:
https://www.donorbox.org/etcg
http://www.silicondojo.com/
“Easy” Computer Vision (Azure, AWS)
“Easy” Computer Vision
-Using Serverless Architecture
-Azure is EASY (60 minutes from account creation was creating scripts)
-AWS Rekognition is “Easy” if you already build for AWS
Silicon Valley is Diseased
-Think about computer vision beyond the dystopia that Silicon Valley is selling.
-Think of correlations vs. identification (uniforms detection vs. face detection)
-Do you need names, ssn, email addresses connected to images?
-The less data you create, the less data you are liable for
-Store Hashes not Images
Algo’s Don’t Replace Management
-Use Vision to Give Insight into business/ organization
-DON’T use vision to trigger decisions
-Be able to explain what the Vision System does, WHY you designed it this way, and WHAT it’s looking for
-Create rules, and enforce rules, not knee jerk reactions
-Have rule change process
General Direction
-Vision is not about being precise, it’s about giving decisions a direction
-“Horseshoes and Hand Grenades”
-Should you show an ad for Tonka Trucks, Barbie Dolls, Ford F150’s or Summer Dresses?
-Self Identification…
--Is it statistically relevant for YOU
--Making the right correlations?
--Nail Polish for Boys
--Female Urination Device
--Serverless Architecture
Demos
-https://azure.microsoft.com/en-us/services/cognitive-services/computer-vision/#features
-Demo General Vision with Azure App
-Demo OCR
Face Identification
-Face comparison
-Color Detection
-Spatial Analysis
-Azure Dashboard
-Setting up Azure
-Code Examples
Azure Issues
-“Public” Images are PUBLIC
-Use Azure Data Storage for “private” images
-Free Service gives 20 requests per minute and 5000 per Month
-Script fails may be because of quota limits
-Step through multiple image recognitions don’t do all at once.
h-ttps://docs.microsoft.com/en-us/azure/cognitive-services/cognitive-services-virtual-networks?tabs=portal
AWS Rekognition
Rekognition Demos