Overview
When searching for a way to upscale my images that I had generated with MidJourney mid 2022 to use as wallpaper, I came across chaiNNer where one can run upscaling models on images. Finding the Upscale Wiki Model Database with more than 200 models with textual describtions but no way to visually compare the outputs of these models against each other, I started building this website for users to visually compare these model outputs for themselves. The newest examples feature around 600 different upscaling model outputs.
Approach
First I compared and chose an appropriate framework (VuePress) I wanted to build this project in. Then I gathered upscaling models, prepared input images, and created the corresponding outputs. From there on I built upon this project: Instead of bundeling the images as assets into the app, I uploaded the images on github, and added a script that would generate a data.json file on build time with each commit on master branch so that the examples generate themselves and i can simply manage the files. I later on added additional categories (deblurring, denoising, etc.) with architectures I ran in google colabs (MAXIM, SCUNet, FBCNN etc.) to produce the corresponding outputs. More info on the Changelog: https://phhofm.github.io/upscale/changelog.html