Approach:

Use Case Discovery:

In this first step, Adastra will outline the business process, potential efficiency gains and general improvements (savings opportunities) of designing, developing and implementing a custom computer vision application for your organization. We will determine image/video features of interest to support the practical business use case, align sample image portfolios and determine anticipated image formats, resolution, and quality for the operational process. Scale and efficiency requirements will be documented to create an operationalization roadmap.

SageMaker Platform Setup:

Next, Adastra will establish a SageMaker environment to enable the development of a prototype AI/ML model, leveraging SageMaker instances

• Set up connectivity to the data store
• Leverage SageMaker Model Registry to catalog and manage model versioning
• Leverage SageMaker Endpoints for model hosting and production
• Enable additional features such as data labelling, preprocessing, model training, evaluation, and production performance to support a target use case

Computer Vision Model Development:

Once the SageMaker environment is set up, custom deep learning models will be used to provide the ability to capture relevant objects and artifacts for your business process. They will be used to:

• Detect objects, resolve duplicates, and track objects frame-by-frame for surveillance use cases
• Identify anomalies and artifacts for condition monitoring
• Detect gaps in operational processes by capturing the number and frequency of relevant objects
• Segment images (if required to determine boundary locations, classify image segments, etc.)

Production Roadmap:

Lastly, Adastra will align orchestration mechanisms to the target use case and scale/efficiency requirements (process SLA), as well as align architecture/services to handle the requirements. A production, monitoring, and maintenance blueprint will be created with a timeline for full-scale deployment.

Activities

• Use case discovery
• Data alignment & annotation (as required)
• Setup of the SageMaker environment
• Data consolidation
• Iterative model development
• Detection post-processing and model refinement
• Results summarization
• Production deployment architecture and roadmap alignment
• Gap analysis to meet expected accuracy of performance requirements (if required)

Deliverables:

Scripts for data collection, model development, and assessment Packaged pipelines for the computer vision model A model demonstration An accuracy results summary A technical workflow summary Production deployment architecture and roadmap Gap analysis (if required)

Outcomes:

• Identification of expected model effectiveness for production deployment
• A gap analysis for any improvement requirements for production usability
• Accurate determination of the investment (capital and operational expenditures) for more productionization
• Architectural alignment for production deployment