SAN FRANCISCO, 10:36 AM, WED JUL 9 | 32 POSTS IN THE LAST 24 HOURS | tips@valleywag.com | RSS
« next »

Image-search startup Riya calls Google's plans "largely impossible"

Shumeet BalujaGoogle-backed researchers Shumeet Baluja (pictured) and Yushi Jing presented the Mountain View company's latest image search and recognition efforts to an audience in Beijing, China on Thursday. VisualRank attempts to do for images what PageRank has done for typical Web pages — rank them in search results according to "authority," which will presumably increase the relevance of results. Problem is, their limited success came at a cost Google is typically loathe to pay: 150 units of homo sapiens who helped sort and rank the images by hand. Munjal Shah, CEO of image-search startup Riya, remarked to the Times: "I think what they're trying to accomplish is largely impossible." Funny, because large-scale, advanced image recognition is what Marissa Mayer says will solve Street View's privacy conundrum.

1:00 PM on Mon Apr 28 2008
By Jackson West
623 views
3 comments

Comments

  • Image of MunjalShah MunjalShah at 04:27 PM on 04/28/08
    Reply by Email

    BTW guys I never said anything about them doing it with 150 people - did you guys get that from somewhere? It is news to me.

    Munjal

  • Image of Owen Thomas Owen Thomas at 07:57 PM on 04/28/08
    Reply by Email
    *

    @MunjalShah: Might want to work on your text-recognition skills. That figure, which came from the Times, was not attributed to you.

  • Image of DanielTunkelang DanielTunkelang at 09:40 PM on 04/29/08
    Reply by Email

    I read the full paper ([www.www2008.org]), and I'm a bit surprised that they're going after product image search for popular products for several reasons:

    1) This is a very narrow case of the general image search problem. To quote the paper, "The categories of queries addressed, products, lends itself well to the type of local feature detectors that we employed to generate the underlying graph." It is, at least, unclear how well the approach performs for other domains.

    2) Product image search seems to call, in general, for returning a single result. They don't report how often the first result is correct. The closest stat they report improving the precision of the top 3 results from 2.19/3 (using a baseline of Google Image Search) to 2.8/3.

    3) The baseline of Google Image Search seems unreasonably naive, since precision would probably be better if the search were limited to (or favored) known shopping sites with consistent structure (like Amazon or Wikipedia).

    In summary, the paper's approach is technically interesting might have more general applicability. But I'll only believe it when I see the evidence, and I find their experiment unconvincing for the reasons enumerated above.

Start a discussion:

Reply by Email

Login with your username and password below. Or comment on this post via email.