17 January, 2022
Song of the Day: I Wanna Get Better - Bleachers
The C++ version is still a flop. I did everything. I switched back to Ubuntu. Nothing works.
On the bright side, I made 3 major improvements. Firstly, I modified the converter algorithm to be faster and more accurtae. I did this after realizing that I could turn the problem into a pidgeon-hole problem. Since I store each character in each frame as a byte, there can only be 2^8 possible configurations of foreground, background, and character values. Knowing this, I generated a list of all possible combinations and estimated their approximate rgb values. Then, for each combination of rgb values, I checked which of the possible combinations was the closest to it and stored that in a file. Now, when the converter runs, it loads that file as a config, and creates a dictionary of those values. Then for each pixel in the given image, the frame conversion is a simple matter of querying the dictionary and writing the byte value to the file. This not only allowed for better color matching, but it also sped up the process by a multiple of 3.
Secondly, I realized that I was being a dumbass. Nothing out of the ordinary. I realized that I was reinventing the wheel, and that it was costing me time. So, intead of resizing the images myself, I used the PIL library, and ended up further increasing my speed by about a multiple of 10.
Lastly, I figured out how to properly use the multiprocessing library to maximize my efficiency.