Corel is a 20-class image classification problem. One of the classes is assigned to be the positive class. The classes are:

  ‘African’, ‘Horses’, ‘Cars’, ‘Beach’, ‘Mountains’, ‘Waterfalls’, ‘Historical’, ‘Food’, ‘Antique’, ‘Buses’, ‘Dogs’, ‘Battleships’, ‘Dinosaurs’, ‘Lizards’, ‘Skiing’, ‘Elephants’, ‘Fashion’, ‘Desserts’, ‘Flowers’, ‘Sunset’


 Original source

Thanks to professor James Wong for permission to distribute this version of the dataset. The data (including the thumbnails of the images) can be found here.

The related publication is:

title={MILES: Multiple-instance learning via embedded instance selection},
author={Chen, Yixin and Bi, Jinbo and Wang, James Z},
journal={Pattern Analysis and Machine Intelligence, IEEE Transactions on},

 Each bag is an image, and the instances are image segments. Each segment is represented by the mean of the 4×4 patch features:

1. three average LUV color components

2. three (sqrt) energy components in the high frequency bands of the wavelet transform

3. three shape components with normalized inertia of order 1,2,3



Files – This file contains a MIL dataset x with 20 different label lists. The default label list is for African, but you can switch to a different version of the dataset by doing (use the labels above):

x = changelablist(x,'Cars');

You need the MIL toolbox to load this version of the dataset correctly. If you do not want to use the toolbox, just load the .MAT file. You can access the data and the label lists by:;

labels=x.nlab;  %Get the k'th label list by labels(:,k)

Leave a Reply

Your email address will not be published. Required fields are marked *