Skip to content
Project Repository for Creative Applications of Deep Learning with Tensorflow course on Kadenze.com
Jupyter Notebook Python
Branch: master
Clone or download

Latest commit

Fetching latest commit…
Cannot retrieve the latest commit at this time.

Files

Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
assets/img
final-project
session-0
session-1
session-2
session-3
session-4
session-5
LICENSE
README.md

README.md

Ansemic Writing with Tensorflow

All my files for Creative Applications of Deep Learning with Tensorflow, an online course on Kadenze.

Overview

The Kadenze course "creative applications of deep learning in TensorFlow" is a great MOOC to learn the TensorFlow framework and manipulate fascinating neural network architectures, which has been a great opportunity to develop my burgeoning creative skills.

The following manifold has been generated during the course with the dataset of my choice (latent space):

Galaxy Zoo manifold

The Galaxy Zoo dataset is not expressive enough, so the visual and artistic impact is not as good as expected. Nevertheless, you can perfectly observe relevant variations of galaxy shape, sharpness, orientation, existence of stars in the field, etc.

For the project, I came to play around with generative hand writing (see below for full credits on inspiration and external code use). After a training on human hand writing, the code is able to generate a kind of ansemic writing:

Generative hand writing

Installation and setup

On OS X, download the Anaconda installation script for Python 3.5 and follow the Continuum instructions. Then, install TensorFlow:

$ conda create -n tensorflow python=3.5
$ source activate tensorflow
(tensorflow)$ conda install ipykernel
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.11.0rc1-py3-none-any.whl
(tensorflow)$ pip install --upgrade --ignore-installed $TF_BINARY_URL

Credits and license

  • the main credits go to Parag Mital and Kadenze, all materials are distributed under the Apache license v2.0 and the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license
  • the assignments come from the CADL GitHub repository, you can find in this repository amazing code, like a ready-to-use VAE/GAN net (strong computing power will be required to train it by yourself)
  • the course heavily relies on the CelebFaces Attributes Dataset
You can’t perform that action at this time.