An openframeworks project that creates a grid of images from a t-SNE projection of ImageNet descriptors of those images.
Run ./setup.sh to download the pretrained ImageNet classifier used to generate the descriptiors.
Reads settings from the data/settings.json file:
{"image_sets" :
[
{"directory":"cool_pics/"},
{"directory": "nice_pics/"}
],
"n_images" : 20,
"draw_tsne": false,
"dims": 2,
"perplexity": 35,
"theta": 0.5,
"normalize": true}
image_sets is a list of directories relative to the settings.json file that contain the images you want to cluster (sub-directories within these are scanned too).
n_images specifices how many images to pick from each image set.
draw_tsne specifies whether to draw the in-progress t-sne to the screen (set to true if you love hypercubes)
Saves a JSON file containing the descriptors in <image_set_directory>/features_4096.json.
I highly reccomend that you run from the console to see logging output (cd tSNE_images_gridDebug.app/Contents/MacOS/, ./tSNE_images_gridDebug)
Uses: