Computer Vision

I am part of a diverse team of researchers from the University of Berkeley, Penn State University and UCL working on slum identification from space using satellite images. This is a pilot project, supported by the World Bank and aims at developing methods to establish alternative, cost effective, ways to establish the boundaries of slum areas without having to resort to land surveys.

The computer vision task has been formulated as a pixel level image classification task and is being tackled with a computationally efficient variation of the U-Net neural network architecture.

Computer vision, slum detection, satelite images, image segmentation, U-Nets, Theano, Pytorch