My reality was once consumed by the ever-watching presence of Snapchat
It didn’t happen,
it wasn’t important,
if it wasn’t saved to memories.
I have been on Snapchat since at least 2014. In early 2021 I decided to break the cycle I was going to be free from the complete digitalization of my reality. The memories range from 07-2016 to 05-2021.
The complete collection of my Snapchat data lives on my computer, extracted from a convoluted zipped folder of all my data downloaded from Snap with javascript.
The app remains deleted off my phone, my account as gone as it will ever be.
As an intrinsic part of my constructed reality, my desire for a representation of my online self once existed most perfectly in this space. Saved were all the stories I had posted to the world, along with personal photos I desired to remember.
All important...
Could I use this data to better understand how the computer saw me?
Could this data help me better understand myself?
I sought to create new memories based on my experiences of the past. Here I could use the past to understand my future.
Training a Generative Adversarial Network (GAN) on this data would allow me to generate completely new memories and thus a better understanding, at least so I hoped.
I augmented my memories into pixelated blocks of colors, even at this moment, the pictures started to move away from my conventional understanding of myself.
I ran the model on the augmented Snapchat data
I received pixelated masses of colors
Bright and neon
What could I discern from this, was my world online bright and colorful?
And I ran the model with more pictures from my full catalog of Google photos along with my Snap data in hopes of providing the network more of what it wanted to better understand who I was...
With all my images blobs began to appear but my self remained elusive
I was aware the lack of labels and large variety of image types caused confusion for the network
There weren’t just images of me
There weren't just images of people
There where images of cats, dogs, pretty carpet, endless remnants of random objects from the era of tumblr aesthetics.
Maybe I could seek some clarity by trying to find images in my mind
Or maybe I created my own personal glitch, from the colors of my created reality.
All the best,
Tess Eschebach
Sidenote on GANs: Steps for (this) GAN include:
- augmenting training images for easier processing
- Running the data through a network to generate a new image
- Checking how close an output image is to the training images
- Reporting the images in a visual manner
- Iterating over these steps many, many times
The GAN is inspired by https://github.com/V-Sher/GANs_Pytorch