Amazon Rekognition - Key Concepts
Ref: https://learn.cantrill.io/courses/1820301/lectures/42157339
Diagram: https://github.com/acantril/aws-sa-associate-saac03/blob/main/2300-MACHINE_LEARNING/00_LEARNINGAIDS/AmazonRekognition.png
- 🔧 Computer Vision (CV) → Deep Learning (DL) image and video analysis service
- Identifies objects, people, text, activities… in images/videos
- Pay-per-use: per image or per minute (video) pricing
- Features
- Event-driven (e.g. integrate with an image uploaded to S3)
- Can analyze live video streams (e.g. Kinesis Video Streams)
- Use cases
- Face detection, analysis (e.g. gender, emotions…) and comparison
- Pathing (movement of objects/people, e.g. in sports)
- Content moderation (identify if something is safe or not, potentially filter it)
Amazon Rekognition - Custom Labels
Ref: https://www.udemy.com/course/aws-ai-practitioner-certified/learn/lecture/44887649
- 🔧 Train Rekognition to identify your custom images
- Label your training images and upload them to Rekognition
- 💡 Only needs a few hundred images or less
- Rekognition creates a custom model on your images set
- New subsequent images will be categorized the custom way you have defined
- Use cases
- Find your logo in social media posts
- Identify your products on store shelves
Amazon Rekognition - Content Moderation
Ref: https://www.udemy.com/course/aws-ai-practitioner-certified/learn/lecture/44887649
- 🔧 Automatically detect inappropriate, unwanted, or offensive content
- Can reduce human review to 1-5% of total content volume
- Can be enhanced with Custom Moderation Adaptors (labeled set of images to train Rekognition on unwanted content)
- Integrated with Amazon Augmented AI (Amazon A2I) for human review
- Use cases
- Filter out harmful images in social media
- Prevent GenAI users from generating offensive images with a GenAI model
