Back to projects
PrototypeMachine LearningTeam of 5 · AI lead
Beninese Food Detection
Led the full AI component: YOLO fine-tuned on over 1,000 manually labelled images to detect the components of traditional dishes. Advanced data augmentation, mobile-optimised deployment, extensible to other African cuisines.
75 %
Average detection accuracy
25+
Food classes detected
1,000+
Manually labelled images
Stack
PythonYOLOPyTorchOpenCVHugging FaceFastAPIDocker
- Manual collection and labelling of a large dataset of local dish images
- Handling the visual variability of traditional dishes (presentation, lighting, angle)
- Optimising the model to run on resource-constrained mobile devices
- Balancing the dataset to avoid detection biases
- Collaborative labelling pipeline with the team
- Data augmentation techniques to enrich the dataset
- Post-processing to improve detection accuracy
- Model capable of detecting over 25 types of local food
- 75% average detection accuracy on the test set
- Solid foundation for extension to other African cuisines
A question about this project?
Contact me