View on GitHub

DATA310

Repository for DATA 310 Summer 2021

For the first image of this project I decided to go with a sea turtle mashed with the Starry Night painting. I chose the turtle because they are my favorite animal and I chose starry night because I thought the blue aesthetic would nicely match the turtle. The first image below shows what the two look like on their own. Next is the gradients found for the turtle by the code. Last is the actual image the code produced.
The first method used a preloaded model from tensorflow_hub. This didn’t take any real alteration on our end. The next method uses a VGG19, a very common neural net that is pre-trained. This model attempts to find different edges. It can then find total subjects, in this case a turtle floating in the sea. This allows the image to keep that object in the forefront rather than being overwhelmed by the art being applied to it. The model has 5 convolutional blocks. We then ran gradient descent to decrease the cost function. This produced the second image combined image. The next method was just rerunning the optimization, this time with a total_variation loss function. This produced the final image. We applied this method in the same way for both images.

Reflecting on this image, I would say that it is representative of the calmer side of the jump start program. Although there has been a lot of work in such a short amount of time, it has felt like we’re just surrounded in useful knowledge, just gliding through it. This image captures that wonder and allure of being surrounded by brilliant minds and ideas, soaking up all the knowledge we can in such a short amount of time. These experiences will be what I remember from the program rather than whatever acute stress it may have caused.

For my second go around I chose an image of a panda and an abstract blue artwork. I chose the panda because it’s my second favorite animal, very creative I know. The artwork is just an image I found by looking up “blue abstract art” because I wanted the blue throughline with the other image. The outputs are in the same order as before: normal images, gradients, and combined image.

Reflecting on this image, it’s representative of the exhaustion following a tough week of assignments. The panda’s slumped position exhibits the feeling of near collapse after you spend so much time on assignments for each class. I imagine that following the final week of the program, including a final speech, final essay and exam for Philosophy, and the final project for this class I will be in a similar position as the panda in this image. It will all have been worth it, however.

I would say that my favorite of the two images is the turtle with the starry night. I like the way it came out and I think the two images blend nicely together, providing a more pleasant image. It gives a more relaxing energy which I appreciate.