ADC2 winners announced

The ADC2 winners were announced last week. Alas, the version of Babelsnap I had submitted only made the top 20 in the travel category. I have learned that I better start work on my ADC3 entry now, as a certain level of polish is necessary to win/place/show, and that last mile takes some time.

I would speculate that another ADC will not occur. The ball seems to be rolling now with Android showing up on devices, and thus development for android should follow suit. If this is true, hosting ADC3 would seem somewhat of a distraction, let alone the extra effort made by google employees to make ADC fly.

But I'd love to see an ADC3.

ADC Round 2

It appears that the ADC version of BabelSnap made it through to the second round of ADC2.

ADC2 BabelSnap! locations

So a few days ago we heard that the first round of ADC2 judging is done, as there have been sufficient amount of scores submitted for all the submitted apps. http://android-developers.blogspot.com/2009/10/adc-2-round-1-scoring-complete.html

I can confirm that I had over 100 users try out BabelSnap. Here is a map of the ADC2 activity with BabelSnap.

I can surmise the kind of scores that BabelSnap v1.0 probably received after I reviewed a small sample of the OCR logs on the server. Most of the OCR results did not yield good results matching what the image contained. And I would expect a person who is busily looking at a sequence of apps to judge would simply mark the score low and move on to the next app.

Hmm, I am curious how these images would have done with the v1.1 release. Maybe I'll look closer at the logged OCR images and make another post in my infinite spare time...

BabelSnap! v1.1 is now on Android Market

Today I released BableSnap! v1.1, the first version available to the public on Android Market.

The biggest shortcoming in the initial version (which was submitted only to ADC2) was that capturing text was not "casual" enough. You had to really work at framing the subject with care in order to eliminate unnecessary content along the picture edges, thereby helping the OCR isolate exactly what you want.

The v1.1 release improves the task of capturing the text you want. You may capture an image without worrying about the presence of extraneous text or graphics surrounding the text of interest. After the image is captured, you use your finger to drag across the text you are interested in, which tells the OCR specifically what to scan.

The success rate of the OCR has improved dramatically with this single change. But now I can see the actual limitations of the OCR itself, and the mistakes it makes. It is far from perfect. But now I have the next list of improvements to go after.

A few other features and bug fixes were made in this release. There are also contrast/brightness controls, and guiding hints for the first time user. The full release notes are found here: http://groups.google.com/group/babelsnap/web/change-log

BabelSnap has been submitted to ADC2

Last week I submitted BabelSnap to google's ADC 2, under the travel category.

I have let friends and family try it out, and it's main shortcoming is that is actually takes some practice to get the hang of it. You have to hold the camera still while it is focusing, frame the text just right, and hit the shutter button -- all at once.

But once you've trained yourself in doing this, you can get good results.

Meanwhile, I am working on the next revision, trying to permit more "casual" snapping of text. One should not need to learn to do a balancing act in order to get an image worthy of sending through the OCR. In a perfect world, the OCR would just magically know what the user wants to keep and ignore in the image. But this is really hard. So in the meantime I'd rather let the user casually snap their image, then allow them to "refine" what should be important to the OCR.

More to come.