As a freshman, my first research-paper implementation (as part of the AI Club of IIT Madras) was Pix2Pix, a conditional GAN for image-to-image translation.
We read the paper Image-to-Image Translation with Conditional Adversarial Networks By Isola Et al. And built and trained the full pipeline from scratch (a U-Net generator paired with a PatchGAN discriminator) reproducing the core results from the original work.
We then extended the model into a practical demo that converts real images into cartoon-style outputs through an interactive application.
Learned: How to read and implement a research paper, PyTorch, Kaggle, streamlit (for app interface), and that nervous feeling when you watch your model train/learn :)
Some screenshots from training (on Kaggle P100 GPU)
