Image Analytics: Curalate

With the steady and building drive towards social media networks like Pinterest and Instagram you would think that image based analytics applications would be everywhere, and you would be wrong.  So far the only image analytics company out there is Curalate – and it doesn’t even position itself as a pure play image analytics provider.  From the website: “Curalate is the most comprehensive platform for marketing with imagery.”

Of course a big part of marketing with imagery is understanding what is going on with your pictures – from how many times they are viewed, pinned and forwarded, to lower level issues like de’duping and aggregating hits for shots of the same product. Beyond that, Curalate seems to have a very close relationship with Pinterest and has recently added management for Tumblr, Instagram and Facebook. One stand out feature is Curalate’s ability to manage Pinterest contests – something that usually requires a disjointed “pin this with a hashtag” or “email a submission outside of Pinterest” to accomplish.

The good folks at Curalate are not saying much about how they do the large scale image recognition work.  In a joking blog post from early 2013 they attribute the image recognition to a squirrel named Shippy.  You would think that a big data, machine learning company like this would probably be based in Boston, but again, you would be wrong.  Curalate is in the tech startup hot spot of Philadelphia.

Shippy, Director of Image Matching

Far from joking around, the other blog posts on Curalate’s blog reveal a powerful ability to analyze social trends around image marketing.  A post titled, “Beautify Your Content: 8 Image Features that Shine on Pinterest” features an infographic that quickly made its way around the web.  It summarized these findings:

People like colorful images, but moderation matters.

  • The most repinned images have multiple colors: Images with multiple dominant colors have 3.25 times more repins per image than images with a single dominant color.
  • Very light and very dark images are not repinned as often. In fact, the repinning rate for images of medium lightness is 20 times higher than for images that are mostly black, and eight times higher than images that are mostly white.
  • Images in the middle between highly desaturated (e.g. black and white) and highly saturated perform better. Images that are 50 percent saturated have four times more repins than images that are 100 percent saturated, and 10 times more repins than images that are totally desaturated.
  • Red images get more repins than blue images. What’s more, images that are red, orange and brown receive roughly twice as many repins than images that are blue.

Get close and personal – people pay attention to details.

  • Images that contain less than 30 percent background (e.g. whitespace) are repinned the most.
  • Repins drop off by 2X to 4X for images containing 40 percent or more of background area.
  • Brand images without faces receive 23 percent more repins. Less than 1/5 of images on Pinterest today have the presence of faces.

It will be interesting to watch this space and see if Curalate picks up competition from other big data machine learning companies or if it can remain out in front long enough to become too powerful to be challenged.