My implementation of Google's Deepdream deep learning image processing software.
This Deepdream repository builds upon the standard Google Deepdream interface to offer a simple, intuitive and easy-to-use command-line interface for generating custom "dream" images.
After sorting through all of the 80+ end layers that are available, I've isolated the 9 most interesting effects and presented them as options for customizing your output. Each end layer offers a unique and striking visual effect to generate everything from streaky images in the impressionist style to the classic dogs-everywhere look that made Deepdream famous in the first place.
There are other editing options avaiable at the command line as well including the octaves (number of passes) and the scale of each octave.
Full alpha transparency support for PNGs has also been built in. Supports JPG and PNG images as inputs, and will always output an alpha-enabled PNG image that can be converted to a flat or lossy format (such as JPG) by the user afterwards.
The modular design also allows users to add custom effect modules and have a little fun with low-level image processing. The alpha support module for PNGs can serve as an example of this functionality.
Run with python main.py or as an executable ./main.py and go from there. All input images must be located in the inputs folder.
There are significant dependencies to run Deepdream. Most of them involve the installation of the Caffe Python module, the Deep Learning framework that was implemented in the creation of this piece of software (Disclaimer: this can be painful. Consider yourself warned). In addition, you must have a working distribution of Python 2.7 installed.
You will also need to install a working Deep Learning training set. See the Berkeley Vision and Learning Center's Model Zoo to find a working model.
For more information on dependencies see requirements.txt.