Anomaly Detection via Midlevel Visual Attributes

Algorithm 2

Learning visual attributes and model parameters.

Input: , ,

Output: , ;

(1)  Initialize by randomization

(2) Initialize :

(3) repeat

(4)  Optimize in by block gradient descent

(5)  Train linear SVMs to update , using as the label for feature

    and the th attribute,

(6)  Update :

(7) until Convergence

(8) Return ,