Skip to content

benjamin32561/Generate-Monet-paintings

Repository files navigation

Image-to-Monet-painting

This project is a Deep learning model used to convert images to monet style paintings, the whole project is written in Python and using the Tensorflow 2 library

Installation

Download the repository and run pip install -r requirements.txt to install all the required libraries

Usage

To use the model to generate paintings run the useGenerator.py code. example: python useGenerator.py model/generator.h5 test_images/test1.jpg the first parameter is the location of the generator, the second one is the location of the image to convert.

Model architecture

The generator model is a trained VGG19 model without the Sequential head connected to a not-trained flipped (from end to start) VGG19 model, pretty simular to an Auto Encoder architecture but without the Dense layer in the middle.

Work process

At first I tried to build it using CycleGAN, Som errors accurd so i tried to tackle the problam using simple GAN, it worked fine but didnt add colors to the results, then I returned to the CycleGan and made it work.

Dataset

Got the dataset from Kaggle, I’m Something of a Painter Myself, cleaned it a little bit for my needs, You can find my final dataset here, Its splitted to landskape and paintings already in numpy files, the images are 256 256 3.

Results

All of the presented examples are new photos I found online, They are not even a part of the dataset!

First example

Second example

Third example