see a puffin eat a fish

When I was in Seattle over the summer, I knew that coming back this fall I would have the opportunity to mess around with Rails. I had some upcoming projects and I could either code them up using systems I knew well (.net, php, django, etc.) or take a chance on the steep learning curve that is Rails. I had a couple of Pragmatic books shipped to me in Seattle, read over them on my endless commute to Redmond but never had the chance to sit down and properly code something.

I’ve been back in New York for two weeks. The first was spent figuring out Rails development environments, finally eating a good bagel, biking around the city, you know the important stuff. Starting this past Monday (it’s Friday now), i’ve manage to knock off half of my functional components list for a project. Half, with almost ZERO Rails coding experience as a starting point.

I’ve been able to code on my laptop without a network connection while waiting for street cleaning to be over (thanks Locomotive) and I am in love with migrations. Now I don’t think I know ruby that well yet, all I know is the rails way, well except for the fact that I still have no interest in writing tests. Old habits die hard. I also haven’t tried to do anything too interesting, everything I am doing is cookie cutter stuff. I’ve heard that’s when Rails gets difficult. It’s the Rails way or the highway and sometimes there isn’t a Rails way. Still, Rails is really dope. For example, the fact that a self-join is done through a concept and through the << operator is awesome.

That’s the update for now. Hopefully i’ll have the same attitude a few weeks from now.

After spending a few days in San Fransisco, Anna and I are driving back east. We are currently in Nevada, where even the slots have slots. On the drive out to Seattle, I kept diligent track of our mileage. Once we got to Seattle, I fired up Google’s ‘My Maps’ and put push pins in every location that we were creating a map of our path that I could share with friends and family.

On the way home i’m going to try something different, a real time generative map using YRB’s ZoneTag. I have the ability to report from anywhere with my cell phone, but without some custom hacking I don’t have the ability to report my location. I’ve used ZoneTag to take photos when my digital camera isn’t handy, but lately i’ve noticed that i’m using it more and more to report my location to my Flickr friends.

To create my real time map with photos, all I have to do is ZoneTag more and put all my photos in a set. I then get my map for free. The photos may not be that great, they’re taken with my beat up Nokia, but they are only one piece of the full story I am able to tell. The other is location.

My version of ZoneTag doesn’t allow me to add the photo to a set directly from the phone. So that still needs to be done manually.

This guy has made a lifelogging device for cats! You can even buy one from him. See more photos here. I also saw this as a bookstore today, LOLKATS in RL:

I went to a great talk today by Alan F. Smeaton from Dublin City University on his group’s progress with SenseCam data analysis. The SenseCam is a lifelogging device that can take color photographs and has an accelerometer, light meter as well as an infrared sensor. Its recording methadology is technology driven versus based on a “user pull” model. User pull involves a user explicitly choosing to document an event, say a concert or a trip to the zoo, while the SenseCam records everything, leaving the determination of the content’s value to the analysis stage. This is the exact sort of data analysis that we wanted to do with the WayMarkr.

Seamton’s group is taking a more computational approach to data analysis, attempting to first analyze SenseCam data through a number of programmatic lenses and then ask the question “is this something?” As an example, he showed findings for a user who had worn the sensecam for 15 months and had over a million photographs in his collection. The user’s millionth photograph was him opening a celebratory box of chocolates, his most important photograph was the first time he met his future girlfriend.

(image analysis using SIFT)

The various methods of analysis are at different stages of development. Some have been published in papers while others do not yet have findings. They involve location correlation using SIFT and other computer vision algorithms, event boundary definition via programmatic segmentation, blob and color analysis, biometric influenced landmark detection (e.g. find me the least blurry photo when my heart rate was the fastest and I was possibly the most excited), event augmentation via public photos (think tagmaps or crowdscapes), really exciting stuff. For the graduate student who wore the SenseCam for 15 months, they found that on average he had 20 segmentable events a day. It would be interesting to compare those values with other populations and it will eventually be possible as SenseCam like functionality moves away from proprietary devices and into more accessible mobile phones.

The team at Dublin City University is even working on concept/semantic feature detection. For example they have trained their systems to detect things like snow, toilet, door, sky and driving. The goal of all this work is to be able to automatically annotate and find representative content in the daily deluge of accumulated content.

On the findings front they currently have three view types for accessing content. A comic book style layout of a day with representative cells of different sizes relative to the uniqueness and duration of the event, a fast playback of an individual’s whole day much like a stop motion video and a map navigation for when location data is available. These views correlate to the four ways people have been found to want to look at their data.

• a ‘gist view’ where a quick glance gives a feel for a whole day
• a ‘relive view’ where we can pan or view a stop motion video
• searching for a specific event like a trip to the zoo
• geographic browsing with google earth to view content relative to where it was collected

Like the WayMarkr, only 4-5% of SenseCam photos were of album quality according to a ground truth survey. The bulk of them were of average quality, but almost all of them were useful at the analysis stage.

I’ve been a big fan of Zurfer from YRB. It’s a low cost way of passively staying in touch with my Flickr friends through visuals on my mobile. Now that school is over and everyone has dispersed, it’s a nice way of seeing where people are in their lives through the Flickr photos they decide to make public. Although my cell data connection is pretty bad, Zurfer is quick enough that I get some good use out of it. I know writing an application like that can be tough since I tried to do something similar earlier this year. I also use Zurfer to annotate stories I tell people in RL. Sometimes a really good story has an accompanying photo for it on Flickr.

I logged into Zurfer yesterday and found myself solicited for user feedback via a temporary Zurfer channel and dummy images (see image). I thought this was a fun and unobtrusive way of asking users for feedback via the core mechanisms of the application. Usually I have a very low threshold for solicitation but the request came to me through the interface that I am using daily. I couldn’t help but say yes.