Research project
Interpretable Image Classification
A final-year research project on classifying images whose classes overlap, in a way people can inspect and understand.
- Client
- University of Moratuwa, Faculty of IT
- Role
- Researcher
- Industry
- Machine learning research
- Status
- Completed
The brief
The challenge
Image classifiers often struggle where classes genuinely overlap, and rarely explain why they chose one label over another. Both problems limit how far people can trust them.
What we did
The solution
The project studied interpretable image classification with overlapping classes, using optimised detection of the overlapping regions so that the model's decisions in the ambiguous areas can be examined, not just accepted. It was presented as a Level 04 research project at the Faculty of Information Technology.
Highlights
What’s inside
- Handles classes that overlap rather than assuming clean boundaries
- Optimised detection of overlapping regions
- Focus on interpretability: decisions that can be inspected
- Presented as a Level 04 research project
Built with
- Python
- Machine learning
Services used
Let’s talk
Want something like this?
Tell us what you want to build, when you need it and what it can cost, and we will say honestly whether we can help.