Projects

Class Imbalance Mitigation with ODOS Tech

Python
PyTorch
Albumentations
Rasterio/GDAL
CUDA
FT-UNetFormer
Knowledge Distillation
Semantic Segmentation

Tackling an industry land-cover mapping challenge using deep learning on satellite imagery, now evolving into a research publication.

Figure 5 from ODOS class imbalance research

This started at the AI Climate Sustainability Sandbox, a five-day challenge at Trinity College Dublin where ODOS Technology provided high-resolution satellite imagery for rural land-cover segmentation — a problem made difficult by severe class imbalance. It’s since grown into a joint research paper (pending submission) targeting a Remote Sensing special issue. I built the experiment workflows and training pipeline.

View GitHub