Project Summary
Motivation
- Ideas orginated from one of takes the digital arts certificate and the other takes the studio art certificate.
Our project consists of two parts
- Artistic Style Transfer by using Neural Algorithm
- Realistic Style Transfer by implementing Deep Photo Transfer
Project Goals:
- Preserve the content of the original input image
- Reference of different illumination, time of day, or weather, or that it has been artistically retouched with a different intent.
- Faithfully transfer the input image into the style of reference
- Implement Neural Algorithm of Artistic Style, Deep Photo Style Transfer
CNN (convolutional neural networks) is a powerful tool on Computer Vision area, especially suitable for image recognition and feature extraction. In our project, we also uses the CNN as our model.
CNN consists of many layers that each contains small computational units of visual information. Each layer can be viewed as a hierarchical-order image filter to the input image, the output contains a feature maps - a filtered version of input image.
Current
State-of-aRT
The algorithm "Neural algorithm" that we are going to use separate the content of our input images and reference style of the images, as it goes down to more image filters, it becomes more care about the actual content of the images. But there are also limitations, for example, the selection of input and reference images. As a result, many people are trying to seek a way to reduce content loss with a faithful style transfer
