the goal of collection is to get a folder of positive (cat head) images and a folder of negative (non-cat) images to train the classifier with.

To get the positives, first download this dataset of cat pictures. There should be folders called CAT_00, CAT_01, etc. Take the images from all of these and combine into one directory. Also remove the file "00000003_019.jpg.cat" and add 00000003_015.jpg.cat.

Run the script to rotate and the crop out the cat head from each image. If you put the cat dataset in a folder called "CATS" and you want to put the cropped images in a folder called "POSITIVES":

node make-positives.js CATS POSITIVES

If you don't already have a bunch of non-cat pictures you can fetch recent images from Flickr and save them in a folder called "FLICKR" by running:

ruby fetch-negatives.rb NEGATIVES

You'll need at least 10,000 images.

To turn the full-sized images into negatives that can be used directly for training or testing, sample them with:

node make-negatives NEGATIVES NEGATIVES_SAMPLED

Where NEGATIVES_SAMPLED is the directory to contain the sampled images.

If you're getting images from Flickr, some will contain cats for sure, so you'll need to weed those out by taking a close look at your hard negatives (see training directory above).