I finished the length-angle classification and wrote out the text files for Sal's classifiers. He will use those for testing soon and should report back with how well we can classify given just the structure features.

In the afternoon, we met as a group to discuss where we are. Sal wanted me to get the images of the length-angle attribute up as soon as possible, which I should be able to finish by the end of the day. Also, I need to make a slight modification to the text files, but that is trivial and can be finished by the end of the day.

Today I finished writing the ppm, but ultimately gif, images for the length-angle attribute. I added that picture to the webpage of structure features and wrote that to a new directory on the CS machines so that we could view the files easily through the web. Also, Sal reported back that the classifiers were correct at a 92% rate! Which is fabulous! I have asked him about meeting with me tomorrow or the next day to discuss how I should approach further clustering/classification.

It seems that Sal isn't available to be seen today. I happened by him in passing, and he said that he would email out some instructions, but I haven't received any word and it is now the end of the day.

I ended up just downloading some articles on spectral clustering and reading a bit in Linda's book, per what chapters were mentioned in Monday's meeting. Hopefully I will be able to get more done tomorrow. If I don't hear from Sal by tomorrow afternoon, I will bug Linda about what I should do next (or possibly ask her in the morning if she is around).

Went to a BISTI meeting today in the afternoon, before planning to meet with or talk to Sal about what to do next. The meeting was about the various interfaces with the functional model database. It was interesting to hear about the various approaches that have been used to access that data--both their features and limitations.

Later on that same day, I heard from Sal about what he would like me to do next. He would like to start running some experiments in which I put together a few different classes of objects by hand, process them with Yi's program, and use my classification to create text files that are then used by his classification program (which uses SVMs). Exactly what classes to use is still a little unclear.

The classes that Sal and Linda would like to use have been clarified to the following: Skyscrapers, residential houses, townhouses. Within those classes, I will have to further subdivide such that each object is in the same relative place and at the same basic scale as the other images in the training and testing sets. I will make a new folder to contain all the classes for this experiment and then just pass Sal the resultant information for classification.