Simplified implementation of Compositional Pattern Producing Network in TensorFlow for the purpose of abstract art generation and for future deep learning work in generative algorithms.
Examples of images generated by the simplified CPPN with tanh activation.
Examples of animations generated by the simplified CPPN interpolating between two divverent z embeddings.
Examples of animations generated by the simplified RPPN gradually increasing the k steps of recursion.
See Otoro's blog post at blog.otoro.net for more details. See my blogpost at w4nderlu.st for details on RPPN.
My contribution:
- added a new model, RPPN (Recursive Pattern Producing Network)
- added a generalized activation function strategy
- added layer norm normalization
- added cosine similary as an alternative linear layer
- porting to Python 3 and TensorFlow 1.0
- mp4 video generation
Requirements:
- TensorFlow 1.0.0+
- imageio for image generation
- ffmpeg for video generation
MIT