Deep Holiday Manipulation
Noah Walsh
Maryland
- 0 Collaborators
Train a deep convolutional neural network classifier on holiday images and use the deep dream technique to manipulate non-holiday images to appear more festive. ...learn more
Project status: Under Development
Overview / Usage
Train a deep convolutional neural network classifier on a custom dataset of holiday images and use the deep dream technique to manipulate non-holiday images to appear more festive. This is the first instance of using a deep dream image manipulation technique on holiday images, and this can lead to improved holiday filters for holidays like Christmas, Hanukkah, New Year's Eve, and others.
Methodology / Approach
Train the Inception V3 network in Keras on a dataset of holiday images such as Christmas Trees, Santa Clauses and Fireworks, and save the weights to be uploaded to a second Inception V3 network and adjust the activations in each layer, the gradient step size, and other parameters to produce a desired output.
Technologies Used
Keras, Tensorflow, Numpy, Pillow, Pandas, Scipy, Python