author: dylatong@amazon.com
This repository is a collection of content to help enable engineers and data scientists to succeed on their Computer Vision projects on AWS.
Examples of modules that you can use together to tailor a workshop for your audience.
I. Object Detection Workshop Package (est. 4 hours)
Below is content you can package up into a Object Detection workshop for SageMaker. You can put together a 4-5 hour agenda with this content.
-
Workshop Guide: Use this as a sample template for the workshop.
-
- Learn to create and manage a quality data set at scale using SageMaker GroundTruth.
- Manage annotation workforces: private, public (Mechanical Turk), and 3rd party vendors.
- Create a labeling job (for Object Detection)
-
Lab2: SageMaker Algorithms- Object Detection:
- Learn to build a custom object detection (Single-shot Detection) from the training data you created in Lab1 without having to write code.
- Learn about hyper-parameter tuning automation.
-
Lab3: Bring Your Own Script- Object Detection:
- Learn how to bring your own script from a deep learning framework.
- In the lab we’ll bring a GluonCV script to train an object detection model (YOLOv3 on mobileNet).
- Learn how to programmatically launch a hyperparameter tuning job, SageMaker local training as well as perform incremental training.
- Learn how to deploy a real-time endpoint for inference.
Progress Journal
05/15/09: Object Detection Module Completed. 3 Labs: Ground Truth, SM Object Detection Algorithm (end-to-end), and GluonCV on SageMaker (end-to-end).
Version 0.1
Journey
SageMaker Path
- Data infrastructure basics
- Data set -- Ground Truth
- Notebook
- Exploration: TBD?
- Algorithms
- SageMaker Algos: SSD, Image Classifier
- BYOS:
- PyTorch, OpenCV
- Git integration
- GluonCV (future)
- PyTorch, OpenCV
- Training
- Distributed Training
- HPO
- Neo-model optimization
- Deployment
- Auto-scaling
- A/B
- Performance monitoring
- Production architectures
- Video streaming cloud inference
- Edge inference- smart cameras.