Week 3

 Unlike the first two weeks of my senior project which saw a large amount of visually impressive progress being made on the image recognition process, my work this week mostly revolved around bug fixing and continual improvement on my existing code. Because fixing bugs and incremental improvement is always a good deal less interesting than getting a process up and running from scratch, this week has also proven to be the most aggravating week so far, even when compared to the compatibility issues I faced during week 1. That said, there's been no decrease in the amount of information I've been consuming, and my knowledge of python package applications continues to grow. Here's a rundown of what's been going on.

Cool stuff I've learned:

-  Applying conditions to your gaussian blur, specifically changing the gaussian kernel, can go a long way for maintaining contour clarity while still reducing visual noise. Super useful when you're working with lower definition binary images like those of DICOM files. 

- Apparently the stack posts informing me that converting from uint16 to uint8 would result in data loss were a total lie, and I've suffered no negative consequences for my use of the comp-sci black arts. 

- open-cvs ability to determine the aspect ratio of a contour can allow for a user to get more reliable shape detection by not having to solely rely on vertex number, ex: determining between squares and rectangles.

- by figuring out a general bound for the area of a given visual target and scaling that for any given image size, target area can easily be included as a third metric for shape detection.

- 3 shape detection metrics are far better than one. 

- my supervisor Dr. Park's parents are from Atlanta Georgia

Annoying stuff:

 - shrinking an image to reduce contour complexity can also result in some contours blending together when dealing with low resolution DICOM files. However, shrinking is necessary because an image that's too large will result in large contours being read as multiple smaller contours. You really can't win. 

-  for some reason, scans that should be for all intents and purposes identical, return different contours. This isn't actually a huge deal since the image recognition still works, but it just bugs me to no end.

- DICOM files are stored in  sets of 40+ separate files, but it's impossible to open any one file individually, since they software that I used to read them (ITK Snap) opens the entire set of 40 + files. Hunting for individual test scans with names that are nearly identical makes my eyes hurt. 

In other news, here're some fun non-code-related things that have happened this week:

-  I finally started working through my library backlog and read Jeff Vandermeer's Annihilation. I started it way too late at night (darn insomnia) and ended up finishing it in one sitting. Definitely among one of the most tightly-paced and most gripping plots I've ever seen in a sci-fi novel. 

- I watched the movie adaptation of Annihilation, which left me with a headache and a sense of confusion - Our neighbor also got a dog :) 🐕 

- I saw fish in tryon creek for the first time in a good long while.

Anyhoo, that's all for now. Until next time :)