Loading...
It's been quite some time in the making, but we're proud to finally announce the launch of Incogna GIS! This division of Incogna is specialized in on-demand geospatial data processing. Just upload your imagery and we let you extract exactly what you need from it, in record time - thanks to our cluster of on-demand GPUs. We spent a good part of this week at Geotec 2010, where our CTO participated in the GIS Cloud Computing round table along with Mano Marks from Google, Chris Pendleton from Microsoft and David Stanley of PCI Geomatics.
Rui Da Silva also presented our collaborative project (Dasymetric Mapping for Profiles of Spatial Determinants of Health Outcomes in Rural Ottawa) that we've been doing with University of Ottawa's Geomatics department. We met lots of interesting folks and are looking forward to seeing how we can help people to extract the knowledge they need from their imagery. If you're so inclined, be sure to watch our GIS site for more interesting stuff!
Posted by Incogna on 2010-04-16
Today at Incogna we were discussing, as we tend to, development on the GPU. It was joked that you almost need the patience of a Buddhist monk in this line of work.
It got me thinking: there's definitely a barrier to entry when it comes to GPU coding. For example, maybe the utility you're using to debug happens to render upside-down. Then, even simple things, like the Arctan2 function, can cause setbacks...I mean, it's helpful when people follow standards, but they don't always do that.
In any case, successful GPU development seems to boil down to some core things:
Impermanence: code is temporary, it will be changed, improved and maybe even discarded (especially in computer vision). See through the specifics, generalize and don't get too attached to a specific methodology or solution. Then again, don't forget to get it working on time!
Unsatisfactoriness: many of us want perfection. But this is often impossible in the real world, especially related to classification systems, and doubly so in computer vision. Even human classification is imperfect, when done on a large scale. You'll always find misclassifications, the real question is: what's really required? 95%? 99%? 99.99999%? Remember, nothing in the psychological or physical realm can bring deep and lasting satisfaction.
Non-self: foster a mind which is not at the mercy of stimuli or moods. When confronted with observations or problems, have equanimity. Do not be confused, see only what is seen. The bug may be hiding in your mental projections, elaborations or assumptions.
Many people may see SIMD computation as confusing & difficult, which in turn causes suffering. But really, this is the nature of our visual world. The cells in your brain (and all brains) work in parallel. Cells tend to do the same/similar operations as their neighbours, and have developed the ability to work together and communicate with each other. So must our algorithms. This is the future of computing.
In many cases, oddly enough, it's often easier to debug computer vision algorithms in SIMD, once you know how.
Be the pixel. Be the feature. Be the object. Be one with the index.
Posted by Incogna on 2009-12-12
We're proud to announce the latest addition to the Incogna family, Shopachu! Shopachu is a new breed of visual shopping site, we've combined Incogna's search capabilities with the ability to combine products together to create outfits, as well as a handy price history viewer with sale notifications. Never again will you be left out of the loop when top quality fashion goes on sale!
Shopachu has been in the works for some time, but we wanted to make sure we got it just right. In a space as crowded as online shopping, a launch like Shopachu needs some serious thought. What did we do? Well, we found some fashionistas. Then, we took a solid look around at what's out there in the online shopping world and what's missing. Finally, we developed Shopachu to fill the void
What's our goal? We're aiming at bridging the gap between pure search shopping and art/shopping. Finding the perfect pair of shoes is great, but what do they go with? And why use your imagination when you can just use Shopachu, put the things you want together, and see exactly how well they work? Give it a try!
Also, why doesn't everyone track product price histories? You'd be surprised at the zany price trends we've been seeing.
In any case, enjoy!
Posted by Shopachu on 2009-09-21
The web is abuzz about Google's latest Similar Images Labs release. I love that Google is finally devoting some resources to this problem, since their main image search engine basically hasn't changed much since it was launched (other than expanding in size). I always get the question: "What if Google is working on Image Search?" and I always answer: "Oh, they are." It's great to finally see the results!
So, a bit of a breakdown: they've taken an approach with a very strong emphasis on metadata and mixed in some colour similarity (with a hint of shape too). The metadata gives the engine a strong basis upon which (I would argue) they seem to have focused mostly on colour content. I've personally never really been convinced that using colour for recognition is a great way to go. My reasoning is simple enough: colour blind people can recognize basically anything a normal person can (save perhaps for the lime/lemon debate). I think that one of the trappings is that colour is easy to approach from a statistical viewpoint, but in reality, it really gives very little discerning power. What's the problem? Well, you sometimes click on a forest picture and you get horses instead. Although their metadata covers up a lot of those problems. That's not to say that colour isn't important, it's a great way to help restrict your search.
In any case, I have to say: I love clicking around visual search engines!
In other news, Incogna has been advancing on a number of fronts: we're doing some Geographic Information Systems (GIS) pilot projects with a number of companies right now. We're effectively bringing the Cloud Computing model to GIS, where the current state-of-the-art is, shall we say more "desktop oriented". We've given some live GIS processing demos (think Google Maps + interactive processing & classification), I think they're pretty cool. More on that later.
In the meantime, there'll be some alpha invites going out soon for our next project (not GIS). Don't want to spoil the surprise, so you'll just have to wait. Just sign up for an Incogna account (if you haven't already) and we'll send out invites it's ready.
Posted by Kris Woodbeck on 2009-04-28
Monetization is a key question for many startups with cool technology. As you may (or may not) know, Incogna was the first company to combine both image and product search results together visually. This has granted us a new way to monetize image search, without displaying any annoying text ads. So, if you search for boots or jeans and some products happen match your query, then we display those products as "Sponsored Images" on the side. This is basically identical to how text search was monetized, only it's been a heck of a lot more complicated to do with images.
Back when Incogna was little more than a sketch on the back of a napkin, a lot of people put forth the argument that visual advertising (i.e. placing product ads within image search) would not work. I was told: "The click-through rate (CTR) will be terrible!" and similar sorts of things. Well, after the launch, you good folks searched and clicked your way around...we analyzed the data and are happy to announce that sponsored images are quite effective! I can't give any exact percentages (since the competition is out there), but let's just say that our CTR gives text search a run for its money. (For a bit of industry background, Google's CTR is usually around 15-20 percent)
So what's the point? Well, some engines tell us they're handling hundreds of millions of image searches every day. If we punch in the numbers, monetizing this level of traffic with a visual advertising system (such as Incogna's) would net a handsome sum, and all without annoying users! With visual search companies jumping up all over the place, effective monetization will be a key differentiator. Sponsored images, they're going to be all the rage.
As for Incogna's plans for the upcoming year, our index is growing every day and we'll be sending out some ShapeShopper invites soon to everyone who's signed up. What else is in the works? We're working on something that's tentatively called the Visual Thesaurus, and there's also plans to process the world's tourist attractions and do something cool with those. So keep an eye out!
Posted by Kris Woodbeck on 2009-01-08