Objective

This project is to extract Content Targets and Style Targets, and to search for an optimal input image that produces similar response as the content and style targets.

Extract Content Target and Style Target

Calculate Loss and Gradients

Results

Content ImageStyle ImageTransferred Result

Different Number of Iterations

Different Combinations of Style Targets

Efficiency

Mainly related to the size of content image, the number of iterations, and the layers in style targets. For 640 x 427 content images, it took around 1000s for 32 iterations, and around 3hr for 512 iterations. With only one layer in style target, it took around 400s for 32 iterations.