This repository contains all resources and files created during the Hackbyte event, jointly organized by Loblaws and Microsoft.
- Event Photos: A collection of moments from the event.
- Sundeep Pothula - Specialized in image labeling and modeling with Azure.
- Mubtaseem Zaman - Core backend development and deployment expert.
- Amit Singh - Focused on image labeling and modeling with Azure.
- Novina Wong - Front End Developer and UI/UX Designer.
- Weqi Liu - Front End Developer.
- Brainstorming and ideation sessions.
- Client communication to identify key challenges.
- Formulating the best plan with team feedback.
- Problem documentation (Google Drive).
- Strategic planning for task distribution.
- Developed a front-end web app, hosted on GitHub Pages.
- Creating a synthetic dataset and manually labeling it with Microsoft Azure Custom Vision API.
- Dataset, event, and team photos (Google Drive).
- Developed five versions of the model with the synthetically created and manually labeled data using Azure Custom Vision API.
- Successful deployment on Microsoft Custom Vision API.
- Continued dataset building by capturing and labeling product images.
- Impressive model performance after seven iterations.
The model showed outstanding results, as evidenced by the following metrics:
| Tag | Precision | Recall | A.P | Image Count |
|---|---|---|---|---|
| Chocomini | 96.8% | 81.1% | 84.4 | 100 |
| Vegetable Baby Food | 94.7% | 72.0% | 86.9 | 110 |
| Apple Baby Food | 94.6% | 62.5% | 82.0 | 153 |
Our solution offers multiple benefits:
- Streamlines the process for in-store shoppers and pickers.
- Provides real-time item availability from the latest shelf images.
- Eliminates the need for extensive camera installations by leveraging crowd-sourced data.
This innovative approach simplifies the process of locating and checking item availability, utilizing advanced computer vision techniques and Azure Custom Vision.