Neural_Style_Transfer

About

Style transfer is generating an image with the same "content" as a base image, but with the "style" of a different picture (typically artistic). This code provides a TensorFlow implementation and pretrained models for Artistic Neural Style Transfer, as described in the paper A Neural Algorithm of Artistic Style by Leon A. Gatys, Alexander S. Ecker, Matthias Bethge.

Folder Structure:

.
├── #1report     - report of the implementation
├── #2nueral_style_transform.ipynb                - implentation using pretrainged vgg16 using imagenet
├── #3vgg.ipynb       - implentation using vgg16 trained with your dataset
├── README.md

VGG structure

VGG16 is a convolution neural net (CNN ) architecture :

1. 2 x convolution layer of 64 channel of 3x3 kernal and same padding
2. 1 x maxpool layer of 2x2 pool size and stride 2x2
3. 2 x convolution layer of 128 channel of 3x3 kernal and same padding
4. 1 x maxpool layer of 2x2 pool size and stride 2x2
5. 3 x convolution layer of 256 channel of 3x3 kernal and same padding
6. 1 x maxpool layer of 2x2 pool size and stride 2x2
7. 3 x convolution layer of 512 channel of 3x3 kernal and same padding
8. 1 x maxpool layer of 2x2 pool size and stride 2x2
9. 1 x Dense layer of 4096 units
10. 1 x Dense layer of 4096 units
11. 1 x Dense Softmax layer of 2 units

Dataset:

Dataset it's from kaggle best artwork all time It contain best artwork of all time from famous artists

Result

After 1000 epoch the accuracy reached 70 precent

+

=

Output

The next step is that we will implement the algorithm into an api that take capture image of a person and then generate a new style with his/her chosen style and print it. For any information contact nadaSamir@student.aast.edu or nourmorse21@gmail.com.