This Face Super-Resolution model from the Korea Advanced Institute of Science and Technology (KAIST) is meant to improve resolution of low quality facial images. The model takes 16 x 16 pixel pictures of human faces as input and scales the resolution up to 128 x 128 pixels. The results are impressive albeit with some strange artifacts (see below).

@jonathanfly decided to apply this model to something other than faces…Emojis.

Because this model is trained specifically to look for facial landmarks it will take any excuse to draw eyes and nostrils on a pixel…These samples are cherry picked and many outputs were not very interesting. I also went to some lengths to encourage the model to make aggressive guesses about facial features.

The results range from charming and hilarious to somewhat horrifying. The examples below include a haunting ghost and lip-peroni. Checkout the post on iforcedabot.com or implement it yourself with the Colab code from @jonathanfly.