Visual search and image recognition is one of the holy grails of consumer Internet technologies. Picitup is jumping into the deep end of this space by announcing the launch of its public beta.
Unlike Like.com (formerly Riya) which focuses on likeness, Picitup focuses on attaining matching images. This differentiation is important as it sets the company on a completely different trajectory in terms of both offering quantifiable value to users, as well as delivering a business model at the end of the day.
An image search on Picitup begins with a textual search actually queried on Google or Yahoo. Picitup will display a set of results only from one of the two—the basis of the decision is the speed and quality of the results. The user can then select which image Picitup should fetch similar images for, or filter the results by Faces, Products, Landscapes and Color. The analysis is made in real time and is based on 100+ parameters including a propriety color space the company developed.
Erick recently wrote that:
It’s hard to compete in the search engine market, but one approach taken by several startups is to sit on top of the big search engines and try to improve their results or interface. Why reinvent the wheel when you can simply add new spokes?
From a practicality point-of-view, relying on the likes of Google and Yahoo makes sense, but it should be noted that they forbid the reordering of their results, a sticking point that surely has a negative effect on the quality of results Picitup ultimately delivers.
Picitup claims it shortens the number of pages needed for an image search from 10 to 2. However, from my experimentation its engine’s match reliability was shaky. Results were pretty good for Ford Focus, but not even close for this Running Shoe. Note that all images should theoretically correspond to the top left-hand image.
Another issue that left a sour taste in my mouth was CelebrityMatchUp, an attempt to add some light-hearted fun to the beta interface. The idea here is that users upload photos of individuals and have Picitup produce results of people they resemble. This doesn’t exactly work. For instance, consider that Michael Arrington’s photo brought back results that he resembles both Barack Obama and John McCain. Huh?
Erick Schonfeld’s photo results are also somewhat curious, although the bright side is that Erick’s wife should be delighted to know she married a Kevin Costner look-alike. For a company claiming its forté is in image matching, Picitup should not have opened this door.
Alon Atsmon, co-founder & CEO, believes the company’s technology is compelling enough to drive revenue both from ads and through licensing a white label version of the engine for integration into ecommerce sites.
True, my initial impressions of Picitup are not necessarily positive ones. However, considering Atsmon is a serial entrepreneur I’ll remain optimistic and wait for Picitup to iterate a couple of more times before I cement my judgment.