At the SIGGRAPH 2014 conference on computer graphics and interactive techniques, the researches first presented their algorithm (see the video above) which can be used to do multiple things, including browsing photos by weather tags and altering the weather or seasonal appearance in a picture. By tagging photos with ambient values such as “rain,” “clouds,” “sunny,” “autumn,” etc., the algorithm is able to learn about the specific visual properties of weather conditions, seasons, or moods associated with certain ambient values.
By creating a large database of photos tagged with a set of ambient values, the researchers trained their algorithm not only to select individual pictures with a specific ambient appearance from a photo collection, but also to replicate certain weather conditions in pictures that have been taken at entirely different conditions. For example, the algorithm is able to make a photo taken during a warm summer day look like it was taken in winter or fall, or introduce rainfall or even the appearance of nighttime into a picture taken in broad daylight.
Currently, the algorithm is still under development, but the research team hopes to have a consumer version of the program soon. According to James Hays, assistant professor of Computer Science at Brown University, the goal of the project is to “make image editing easier for non-experts” – that is, casual hobby photographers that aren’t familiar with all the secrets of Photoshop and other professional editing tools.
While the results generated by the algorithm still look a bit crude, there’s hope that the more images will be fed into the database, the more accurately the algorithm will be able to replicate certain weather conditions in photography. Also, as research progresses, changes may be made to how the algorithm applies ambient values to photos in order to make the results look more realistic.
So, does this mean that in the future, we can take a picture at any time of day and our computers will make them look the way we want them? Probably not, at least not too soon. After all, there are limitations as to what the algorithm can do. Where there’s no sunlight or green foliage to begin with, it’s rather difficult to add these things to a picture. But for simple edits to a picture’s appearance, the algorithm developed at Brown University seems to be rather capable already.
- Teaching machines to see illusions may help computer vision get smarter
- Machine learning? Neural networks? Here’s your guide to the many flavors of A.I.
- Get your Sagan on with 60 awe-inspiring photos of the final frontier
- How to download Instagram photos from any device
- Picture this: The Aura packs thousands of photos in a single frame (for a price)