Credit: Adam Smith
Title: Evolution Image Generation Description: This project is mainly run using an evolution-mimic behavior algorithm. The algorithm is given 3 different images, but cannot obtaian any data from them. The goal of the project is for the algorithm to slowly understand what the original picture looks like through the transfer and mutation of genes from parents to children. Explaination of design parameter choices: The original parents are named Adam and Eve. They are both rectangles but carry very opposing traits. I chose to start with 9 different alleles, this way, the generated rectangles can be more mutable. The main genes are color, position, size, and orientation. Each genes have their own alleles that changes the overall design of the rectangle. Explanation of mutation approach: There are 2 different ways that the algorithm can generate babies. First, the algorithm generates new babies through mixing genes from two random parents and alterating them slightly. Second, the algorithm generates new babies through finding the average number in genes from two random parents. Once the generated rectangles are placed, the algorithm compares the new score with the old high score. If the generated image recieves a greater score, the rectangle is kept as a parent. if the generated images recieves a lower score, the rectangle is disposed and a new one is generated. Artist statement + what worked, what didn't: Something that worked was being able to effectively use evolution as a blueprint of the algorithm. I found it absolutely amusing and would love to experiment it more to pursue better results. However, something that still bothers me is how stupidly slow the algorithm takes. I wish that this assignment was not about evolution and more about getting a desired score as fast as possible. Maybe next time, I can develop an algorithm that does not require obtaining data from two different parents, and only adjusting each iteration of generated rectangle slightly to achieve a preciser effect each time.