The goal of this project is to train a neural network that can generate (probably unfunny) captions to the most popular images, within their respective "styles".
Why? Because why not.
This project is written in Python 3.8, using Tensorflow 2.2.
(This is an actual caption generated by the model).
The model uses a pre-trained DistilGPT-2 from huggingface/transformers as a basis for fine-tuning. Based on over 40 most common image templates, it tries to generate a new caption to a submitted template.
My inspiration was an article by Dylan Wenzlau on Medium.
I am using images from imgflip.com.