Lucid Dream is a python script for running Google Deepdream. It uses a slightly-altered version of Bat Country, you can look there for requirements.
Lucid Dream allows full control over dreams, including:
- specify different iterations, scales or layers for each octave
- zoom (with rotation) towards a target
- save full-sized visualizations of the dream process
- try multiple datasets, layers or guide images
There are 3 datasets included:
- The original BVLC ImageNet GoogLeNet,
- The MIT Places dataset,
- The CompCars dataset.
To run dream(s), use lucid.py:
python lucid.py -i --image #input image (required)
-o --output #output filename root (required)
-b --base-model # list of datasets (default: google)
# acceptable entries: google, place, car
-l --layer # list of layers (default: conv2/3x3)
# can also use /all to try whole set of layers
-v --vis # visualization - output images of the dream in progress
-oc --octaves # number of octaves in dream (default: 4)
-it --iterations # number of iterations in dream (default: 20)
-os --octave-scale # scale of octaves (default: 1.4)
-ol --octave-layer # list of layers for each octave
-oi --octave-iterations # list of number of iterations
# for each octave
-d --dreams # number of dreams (default: 1)
-z --zoom # scale of zoom (default: 0)
-c --center # x,y coordinates of zoom target
# (default: center of image)
-r --rotation # rotation of zoom (default:0)
-g --guide-image # list of guide images to tryBetter documentation to come