ImageFlow is a recently published project from Microsoft Research which aims to re-think the nature of image search. Rather than displaying search results in a grid or flat page-like layout, ImageFlow uses a 3D space in which the searcher can explore the content.
The project is built upon some fascinating prior research. Specifically, that user intent can be quite different when they’re searching for images rather than text. For example, image search…
- Can be ‘exploratory’ and ‘goal-directed’. But, it tends to be more exploratory than web search.
- Is often used purely for entertainment to play or explore in visual space with no end goal.
- Encourages tangents. If a person is browsing for entertainment, this kind of tangent is actually desired.
ImageFlow supports these behaviors with some quite novel techniques. The interface starts with your intended search results displayed in 3D space which you are free to navigate. As you explore, new results automatically appear, giving a feeling of dynamism to the content. Related searches are automatically performed, and those search results appear as you move further right and left, giving you a broader range of images to look at, and to “foster serendipitous discoveries.”
It’s great to see experiments like this. And there’s sure plenty of cool ways it could be developed. I asked the authors about what’s next…
The current implementation does not learn from the users’ behavior/interaction history. We think it would be pretty cool to have the system learn not only from the current user, but also leverage the interaction history of other users or nodes in the social network. E.g., use user or social interaction history to rank the related queries, or to fuel an “auto-pilot” mode where the system drives itself based on what others have selected.
No plans just yet for a public beta of the system. But who knows… if there’s enough demand they would consider it.
Thank you ImageFlow team, Varun Jampani, Gonzalo Ramos, and Steven M. Drucker, for answering my questions and sending me the screenshots.
(Link via FastCompany.)