Pepes, can't get enough of them, that's why I made a website that turns rough sketches of pepes into a rare pepe (unique to your sketch!).
The approach for this is very similar to how I did my Bob Ross styled paintings, the only difference is the architecture of the network and the dataset.
The network uses the pix2pix architecture, so if you would like further reading on it, I would highly recommend reading this blog post by Christopher Hesse.
TL;DR version would be that pix2pix is able to translate an image from domain A
to domain B
, in other words a rough pepe outline/sketch into a real pepe.
To translate a rough pepe sketch into a pepe, I needed some data. Luckily for me I found a decently sized dataset of rare pepes. However, some of them were unusable as it contained too much unnecessary information:
So, I wrote a script that aids my in manually cleaning up that dataset. It displays the original input image on the left, and the image with canny edge detection applied on the right.
If its acceptable just press Y
and it saves both of the image into a designated output folder, else press N
.