Project Description

This is an implementation of fast neural style transfer using autoencoders. I used the VGG architecture to measure the style loss of the generated image against that of the original style image. Results can be seen below

File Descriptions

• model.py - Autoencoder model definition
• vgg.py - VGG custom architecture to extract features
• loss.py - Function definitions of Content Loss and Style Loss
• utils.py - Helper functions
• dataset.py - Custom Dataset to provide images during training
• main.ipynb - Main training script

Results

Pre-trained Models

Pre-trained autoencoder models can be found here

Dependencies

• PyTorch 1.4.0
• Python 3.7.6