NeuralStyleTransfer

This project was an assignment for the course "Advance Deep Learning for Computer Graphics" at University of Bonn, Summer Semester 2020

Requirement :

1. PyTorch
2. TorchVision
3. Numpy
4. Scipy

How to Run :

1. Training:

python neural_style/neural_style.py train --dataset </path/to/train-dataset> --style-image </path/to/style/image> --save-model-dir </path/to/save-model/folder> --epochs 2 --cuda 1

2. Evaluation on an Image

python neural_style/neural_style.py eval --content-image </path/to/content/image> --model </path/to/saved/model> --output-image </path/to/output/image> --cuda 0

Parameters Usage

1. epochs - # of epochs to train (2 is more than enough)
2. cuda - 0 for cpu, 1 for gpu usage

Download data for training

Path to download data : http://mscoco.org/dataset/#download. (13 Gb)

Loss Functions

Style Loss & Content Loss while training

Evolution of content image while training

Effect of noise in content image

Effect of clustered and indoor content image