return home.
We are collecting, analyzing, and making available millions of images from webcams around the world.
We demonstrate that it is possible to geolocalize static cameras using natural variations in the world (e.g., the diurnal cycle and the weather)
This work explores techniques and surveillance applications of keeping background models at different timescales.
We learn the distribution of object shapes conditioned on image location and use this to improve moving object segmentation and to enable anomalous shape detection.