ImageFeaturesClassifier

Using indico's imagefeatures API and scikit-learn to produce a solve a novel image classification task with small amounts of training data.

CIFAR-10 Performance (Logistic regression on raw pixels):

100   examples: 19%
500   examples: 24%
1000  examples: 25%

CIFAR-10 Performance (Logistic regression on indico imagefeatures):

100   examples: 45%     
500   examples: 64% 
1000  examples: 67%

Usage

python model.py

In order to use indico's batch imagefeatures API, you'll also have to set your username and password in your ~/.indicorc file.

[auth]
api_key = "YOUR_API_KEY"