Co-hosted by Capgemini and Microsoft | 2021
Sustainability Data Science Challenge
Together with a colleague from university, I built a satellite-imagery segmentation pipeline that estimates CO₂ emissions, biodiversity and solar potential. It placed 1st for model performance and 2nd overall.
- 1st
- for model performance
- 2nd
- overall
- +80%
- performance from custom augmentation
What we built
A colleague from university and I entered as a team. We built a Keras image-segmentation pipeline that reads satellite images and estimates CO₂ emissions, biodiversity and solar potential. The training data was heavily imbalanced, so we designed custom augmentation to rebalance it, which improved performance by 80%.