Custom Vision PoC - a 2 week PoC

We are a Microsoft Azure Gold Partner offering an extensive range of AI and Cloud Services: from implementing custom made machine learning models and AI Cloud Architecture consulting. We are also the fastest growing technology company in 2021 according to the Financial Times. We offer a 2-week project to build Proof of Concept artificial intelligence model for image classification or object detection on images using Microsoft Azure Custom Vision. The aim of the PoC model is to validate quickly hypothesis if computer vision could be used to solve the given business problem with image classification or object detection. The result of the project would contain the AI model ready to be deployed and a report describing the performance of the model and verification if it meets the business goals and success criteria. This could be the first step in your AI transformation.

Sample use cases include:

Image classification:

• Product quality inspection
• E-commerce photo attractiveness
• Product, animal, building type classification Object detection
• Production line product defect detection
• Car monitoring on parking lot
• Counting objects on product shelves

Requirements:

• 500 – 5000 client’s labeled images (labeling is not part of the project)
• Tags / clasess: 2 – 10
• Min labeled images per tag: o Classification: 50 o Object detection: 50
• Accepted images types: jpg, png, bmp, gif
• Min image height/width in pixels: 256
• Max image height/width in pixels: 10 240
• Max image size (trening): 6 MB
• Max number of objects per image in object detection: 300
• Max number of classes per image in classification: 100

Results of the PoC project:

• Validated hypothesis whether a given problem is solvable with deep learning computer vision models
• Trained model on the given task
• Report describing the performance of the model and error analysis
• Recommendation for future development

Project outline: Week 1:

• Data gathering and understanding
• Business problem definition
• Data preparation
• Data upload to Microsoft Azure Custom Vision and validation Week 2:
• Model training
• Error analysis
• Data improvements (optional)
• Report & recommendation generation