The research paper, called “Transient Attributes for High-Level Understanding and Editing of Outdoor Scenes” is set to be presented at this year’s SIGGRAPH, and with the research, a user just has to type in a simple text string like “less winter” or “more night”, and the algorithm will automatically apply those modifications to your photograph.
The way this was achieved was by gathering a huge number of images (more than 8,000), and then codifying them with Mechanical Turk for 40 different characteristics. By examining the differences in images shot from the same position, but at different times of day and year, and with different attributes such as “rainy”, “gloomy”, or “dry”, a machine learning process is able to apply those same modifications to any image.
While the image modification system itself isn’t up and running right now, you can use a web interface for some of the other tools the Brown team developed, for viewing, browsing, and searching the image database.
While the edited images may not hold up under close inspection, and someone with a decent level of Photoshop skill could probably do them manually, it’s an impressive piece of research, and one that’s usable by anyone. So in the future, maybe rather than picking a filter from Instagram, you’ll just tell your smartphone to make the image “more Summery”.
[via Gizmodo Australia]