Computer Vision and Deep Learning (Project)

Superpixel Clustering and Post-Hoc correction to improve the performance of cloud segmentation from satellite images

Abstract:

We implement a 1D-SEUNet model on Superpixel vectors of satelite images to segment out the cloud from the background. Our method aims to reduce the parameters and increase the speed of a conventional segmentation model.

Main Libraries used

• Pytorch
• Fast-SLIC
• Rasterio
• Sklearn (MDS and PCA)

Model

Results