theano_artistic_cnn

Implementation of generating artistic pictures[1] in Theano[2]. We used AlexNet here, even though original work used VGG.

This code is light enough to be run on the CPU only. To enhance convergence speed, RMSprop is employed for learning rate schedules.

Trained weights are taken from [4].

How to run

Prerequisite : Theano[2] and Pylearn2[3]

Run preprocess_img.py : python preprocess_img.py
Run artistic_alexnet_inference.py : THEANO_FLAGS='floatX=float32,device=cpu,nvcc.fastmath=True' python artistic_alexnet_inference.py
Run artistic_alexnet_train.py : THEANO_FLAGS='floatX=float32,device=cpu,nvcc.fastmath=True' python artistic_alexnet_train_rmsprop.py

[1] Gatys, Leon A., Alexander S. Ecker, and Matthias Bethge. "A neural algorithm of artistic style." arXiv preprint arXiv:1508.06576 (2015).

Name		Name	Last commit message	Last commit date
Latest commit History 11 Commits
.ipynb_checkpoints		.ipynb_checkpoints
cnn_features		cnn_features
lib		lib
pretrained_weights		pretrained_weights
results		results
.gitattributes		.gitattributes
README.md		README.md
artistic_alexnet_inference.ipynb		artistic_alexnet_inference.ipynb
artistic_alexnet_inference.py		artistic_alexnet_inference.py
artistic_alexnet_inference_tutorial.ipynb		artistic_alexnet_inference_tutorial.ipynb
artistic_alexnet_train.py		artistic_alexnet_train.py
artistic_alexnet_train_rmsprop.ipynb		artistic_alexnet_train_rmsprop.ipynb
artistic_alexnet_train_rmsprop.py		artistic_alexnet_train_rmsprop.py
artistic_alexnet_train_rmsprop_tutorial.ipynb		artistic_alexnet_train_rmsprop_tutorial.ipynb
kaist_n1.jpg		kaist_n1.jpg
kaist_n1.npy		kaist_n1.npy
preprocess_img.py		preprocess_img.py
preprocessing_img.ipynb		preprocessing_img.ipynb
preprocessing_img_tutorial.ipynb		preprocessing_img_tutorial.ipynb
results.npy		results.npy
setup_nbserver.py		setup_nbserver.py
van_gogh_starry_night.jpg		van_gogh_starry_night.jpg
van_gogh_starry_night.npy		van_gogh_starry_night.npy