Generalization Evaluation employing a Training and a Testing set In order to show the generalization capabilities of the learned color metric, a visual comparison between segmentation results of different images in a same context was performed, were we used a common learned color distance metric to segment them. This experiment employed 60 images of brown horses divided into a training set and a test set. Since there were no GTs available, we followed the segmentation strategy of iterating the algorithm until one of the following stop criteria was achieved: (a) the object of interest, namely the horse, was contained mainly in one segment or (b) a segment leakage into or from another, non-horse object, prevented a better segmentation of the horse. Click on the images to see in High resolution. TRAINING TESTING The complete set of these images and others (imagens and parameters) can be visualized below Training Set Testing Set