---

Durer

Motivation

To create a web app based on WGAN-GP that generates realistic Albrecht Dürer paintings

Requirements

Python 3.8 or above with all requirements dependencies installed. To install run:

$ pip3 install -r requirements.txt

To run non super-res version

$ streamlit run duhrer.py

or check it out here: https://duhrer.herokuapp.com/

To run super-res version

$ streamlit run durer.py

To check out paintings with different super-res manaully

$ python3 check.py

-->

super-res

non super-res

Architecture and other details

Trained for nearly 150 epochs on approximately 8000 Albrecht Dürer paintings

LapSRN_x8 used to upscale paintings by a factor of 8 (pretrained)

w1 ---> weights saved at 100 epochs, w2 ---> weights saved at 150 epochs, total epochs ~150

Partial Dataset ---> https://www.kaggle.com/ikarus777/best-artworks-of-all-time

all images resized to 64 x 64 x 3(channel)

Results

final scores: loss_g: 0.5128, loss_d: 1.1873, real_score: 0.5859, fake_score: 0.0469

Might Do

• Upload WGAN-GP Notebook

---