CloudEater

GEOSS Hackathon demo of a gamifying approach to cloud classification.

Install instructions

git clone https://github.com/mt-krainski/cloudEater.git
cd cloudEater
pip install -r requirements.txt
python newMain.py

Gameplay

• The player controls a "cloud-eating" airplane on a randomly chosen Sentinel-2 satellite image. Their objective is to mark all the clouds in the picture.
• After the player is satisfied with the selection, they submit the cloud marking using W. The player's submission is compared and heuristically scored against other submissions in the database, or the ground-truth data (if known).
• All players' submitted data for each image is collected and can be used to reconstruct the cloud cover using statistical methods.

Player's submission.

Heatmap of previous guesses in the database - bright areas are very likely to be clouds, since many players marked them. Areas with only one or two submissions are most likely outliers.

Scoring of the player - the player receives positive points for green areas, and negative points for red areas.