/Detecting cars in aerial photographs with a hierarchy of deconvolution nets

Detecting cars in aerial photographs with a hierarchy of deconvolution nets

Satyaki Chakraborty, Daniel Maturana and Sebastian Scherer
Tech. Report, CMU-RI-TR-16-60, Robotics Institute, Carnegie Mellon University, November, 2016

Download Publication (PDF)

Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author’s copyright. These works may not be reposted without the explicit permission of the copyright holder.


Detecting cars in large aerial photographs can be quite a challenging task, given that cars in such datasets are often barely visible to the naked human eye. Traditional object detection algorithms fail to perform well when it comes to detecting cars under such circumstances. One would rather use context or exploit spatial relationship between different entities in the scene to narrow down the search space. We aim to do so by looking at different resolutions of the image to process context and focus on promising areas. This is done using a hierarchy of deconvolution networks with each level of the hierarchy trying to predict a heatmap of a certain resolution. We show that our architecture is able to model context implicitly and use it for finer prediction and faster search.

BibTeX Reference
author = {Satyaki Chakraborty and Daniel Maturana and Sebastian Scherer},
title = {Detecting cars in aerial photographs with a hierarchy of deconvolution nets},
year = {2016},
month = {November},
institution = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-16-60},