EN FR
EN FR
RITS - 2018
New Software and Platforms
Bilateral Contracts and Grants with Industry
Bibliography
New Software and Platforms
Bilateral Contracts and Grants with Industry
Bibliography


Section: New Results

Realistic Weather Augmentation for Evaluation of Bad Weather in Computer Vision

Participants : Raoul de Charette, Shirsendu Halder.

Computer vision is evaluated on extensive databases that include large number of examples and allow the ranking of algorithms. However, all databases are acquired in clear weather conditions, where the atmosphere is a transparent medium. In rain/snow/fog, when the atmosphere is filled with particles the light is refracted/reflected/diffracted and the appearance is altered. Here we propose a new research that augment existing databases with new weather or arbitrary amount. We applied it on Kitti and Cityscapes. Our approach uses an accurate understanding of physical and optics models to generate realistic rain/fog and augment existing images or sequences. This allows us to evaluate state-of-the-art vision algorithms for both object detection and semantics and quantitatively measure the effect of rain or fog on them. This research was conducted in collaboration with Jean-Francois Lalonde from Université Laval and was supported by Samuel de Champlain Quebec-France collaboration program.