A follow up to the last post. There have been many examples of “irrational exuberance” about demos at TechCrunch. These are a few – related to image analysis and computer vision – that I can recall.
- Image resizing (below).
- Measuring distances from photos (Visual Size). Based on two or more photos, via stereo vision, so it can work. But there is no prototype! At least the creators were knowledgeable and replied to a couple of skeptical comments.
- 3D models from pictures (don’t recall the name of the company). Based on a single picture, so it can’t possibly work as you need conversion 2D to 3D.
- Raya and its “first ever visual image search engine”. You could actually try it before you get excited. If you did, you would not be as impressed.
- Microsoft’s Photosynth. Clearly, it can work (stereo vision again). But, since you can’t start your own site (location, not web site), we don’t know how much manual work is involved.
So, where does this naiveté come from? I think most of the readers are what we would call “technology enthusiasts”. As a result they are …well… enthusiastic just about anything. And since they create little themselves, there is no “filter”. Another news site that I frequent is YCombinator News (Hacker News). The readership is supposed to be different but just as at TechCrunch what you see is dozens of wows and one or two skeptical posts. The latter are ignored. Business of Software is the best. I like that no-one would even consider posting a half-cooked demo instead of a working software.
At TechCrunch the amount of excitement shown by the posters and commenters is often ridiculous! The latest example the post about is image resizing software. The first 30 comments (and the original post!) are only ah and oh and wow. No-one wants to take a second to think before writing a comment. Isn’t it clear that this can’t possibly work for all pictures? All of the examples have horizontal patterns, so you can shrink them horizontally. If a picture had a vertical pattern, you’d be able to shrink it vertically. What if there is no pattern such as a face? You have to do this manually. What if there are no “seams” they speak of such as in a circular pattern? You are stuck. Just a thought… This is a nice work but it will be forgotten – by this crowd – within a week.
The site is mostly complete now. At least everything is operational. The last additions are: this blog, the image gallery, and the wiki.
The blog is still thin, but at least it’s current. Note that the older posts came from the blog-like page of the old site.
The image gallery has more than 250 images arranged in 9 albums: Medical, Image manipulation, Watercolors, Industrial machine vision, Biometrics, Microscopy, Maps, Test images, and Material Science. They contain examples of image analysis and manipulation with Pixcavator and a few comparisons to other software. There is a need for more – everybody is welcome to contribute.
The wiki has plenty of content now. It started with four chapters from the book draft but now has grown beyond that. There is a thorough explanation of methods, with code snippets for illustration. Now we have added some complete code as well a description of the SDK (in progress). Initially the intent was to concentrate on developers, now it seems to make sense to add articles about how Pixcavator is applied in specific areas of scientific image analysis. Right now microscopy and cell counting seem the most promising.
A couple of articles here and here. In short, the interest is in semantic tagging based on image analysis. The methods are not explained. An example of image analysis in the second article is very optimistic (“hair”, “smile”, “woman” etc). What separates them from the pack is the test they have conducted with the help of large computer grid in the UK. Millions of images instead of thousands. That’s great. Still, why can’t we see a little online application? The collection does not have to be huge to show the capabilities of a search engine. My impression is, it’ll take a while.
I ran into a report by WSJ on LTU TEchnologis and ImageSeeker: ”The company has come up with software that can recognize, index and describe images according to their visual features.” “ImageSeeker [is] an image-comparison engine allowing users to search for photos and videos using images instead of keywords.” I said to myself ”Nothing new” and then I saw the date November 19, 2001! I guess the news is that how little has happened since… To confirm, I went to the web site. The only demo is the one I saw last time I was there. There is a collection of a few dozen images and you can “find similar images” – from the same collection. The search continues…