It is incredible that two weeks have passed since I began this internship. I am already two thirds of the way through! This is quite terrifying because it means I only have four weeks before my final presentation. Hopefully I will have developed a decent app by then so I will be able to give a good demo.
This morning, I spent most of my time working on creating high quality, color images out of the new fits images that Dr. Kartaltepe put on Google Drive for me. Now that I understand how to use the configuration file, I am pretty good at manipulating the values in the file such as color saturation, minimum level, maximum level, and gamma type. By adjusting these values, I can increase the quality of the final color images to make galaxies stand out. I worked on just one image until I made it the best quality I could, and then used its configuration values on the other 20 images. They all turned out very well, but I could see that each individual image requires its own unique adjustments to the configuration file in order to produce the highest quality version of that image.
Today was the weekly lunch lecture, so at 11:30 I went upstairs to eat the free pizza with the rest of the interns. Unfortunately, someone snagged the last grape Fanta before I could grab it. They should really have more than just two cans of the best flavor soda!
The talk was giving by Christopher Kanan, a professor working on visual machine learning and visual question answering. He told us about previous projects he had been working on and the problems that he has faced. A large part of his work revolves around visual question answering (VQA), which is essentially when a computer looks at an image and answers questions about it. The primary issue he faces regarding this work is that when the computer gets the answer wrong, it is very difficult to determine what processes the computer went through to come to its answer. This makes it hard to know what to fix.
In the afternoon, I spent time developing a new programming class representing an individual galaxy. Each galaxy object has total mass, black hole mass, star formation rate, stellar mass, and gas level instance variables. Since I have figured out how to detect collisions between galaxies in the game, now I can work on changing the values of the instance variables according to the type of collision.