The aim of project can be summarized as making a fusion image from 2 different images named as source and style images. However, while doing that, the fusion should have traces from both images in terms of their styles and semantic meanings. When the fusion image is examined, observer can see the markers of two root images. However, in this progress there is no completely new different image. This procedure can be defined as reinterpretation of source of image by using the style image. To achieve this progress, re-implementation of article of "Image Style Transfer Using Convolutional Neu-ral Networks" (Gatys et al., CVPR 2016) was applied. This algorithm was used with different Convolutional Neural Network (CNN) architecture from the paper. SqueezeNet model from “SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size " (Iandola et al., 2016) was chosen due to less computational power requirements. Also, another model Vgg 16, from “Very Deep Convolutional Networks for Large-Scale Image Recognition " (Simonyan and Zisserman ,2014) ,was chosen for comparison. Different Turkish painters with different artistic style such as more realistic, or more surreal were chosen.
- Photo of Informatics Institute obtained from https://ii.metu.edu.tr/tr
- Image of Lenna Forsen obtained from wikipedia
- By Ulpiano Checa,
Le Temps obtained from wikimedia - By Van Gogh,
Starry Nights obtained from wikimedia - By Salvador Dali
The Persistence of Memory obtained from britannica - By Rogier Van Der Weyden,
Portrait of a Lady obtained from wikipedia
Homeworks and its lectures of the course below
Johnson, J. (n.d.). EECS 498-007 / 598-005: Deep Learning for Computer Vision. https://web.eecs.umich.edu/~justincj/teaching/eecs498/WI2022/