This is a Computer Vision Hackathon led by Mckinsey Quantum Black (2023). The goal was to imagine a tech start-up named Foodix leveraging machine learning to detect silos and track their condition in real time.
Tech side of the project was composed of two parts:
-
Image Classification (Silos Detection): using CNN with Adam optimizer and Binary Cross Entropy loss
-
Image Segmentation (Silos Location): using a U-Net CNN
For final presentation, we built a web application using streamlit in order to perform live predictions and localisation of silos
- Image Classification: AUC of 0.93
- Image Segmentation: Dice coefficient of 0.75
- Clone the project
- Install dependencies
pip install -r requirements.txt
- Install Package
pip install -e .
- Run streamlit app
streamlit run src/app.py
N.B: It's also possible to run through Docker following these commands:
docker build -t silos .
docker run silos
- Lucas Chaix
- Simon Mack
- Charles Proye
- Youssef Jouini
- Adrian Tan
- Nathan Aïm