Fox, Tiger, Elephant



There are three datasets, Fox, Tiger and Elephant. The bags are images, and the instances are image segments. For each category, positive bags are images that contain the animal, and negative bags are images that contain other animals (also from other categories, not just from the three categories here).

Original source


  title={Support vector machines for multiple-instance learning},
  author={Andrews, Stuart and Tsochantaridis, Ioannis and Hofmann, Thomas},
  journal={Advances in neural information processing systems},

The extracted features for the data were obtained here

Files– This file contains three different .MAT files for the Fox, Tiger and Elephant problems. You need the MIL toolbox to load this version of the dataset correctly.

4 thoughts on “Fox, Tiger, Elephant

  1. Amine

    Hello, thanks for posting this valuable resources! I wanted to ask about the “Results of MIL classifiers” page, can you share what validation method was used to generate the results? Cross Validation or leave one out?

    Thanks in advance!

    1. Veronika Post author


      It is 10-fold cross-validation repeated 5 times. But for some dataset/classifier combinations, due to the run time less folds were used.

  2. Jose F. Ruiz

    Hi! I wonder what does parameter of the milproxm function “INSTPROXM” mean? I guess is the proximity function at instance level. If so, what is the default function and how can I modify it? Thanks!

    1. Veronika Post author

      Sorry for the late response, your comment was in the spam 🙁
      Yes, you are right – it is meant to be the proximity at instance level. By default this is squared Euclidean distance. In the current version it is not possible to pass your own parameter here, but you could modify the milproxm.m code by replacing sqeucldistm.m with another function. For instance-level proximities, you can use regular proxm.m which has several options, like Minkowski, RBF and others.


Leave a Reply

Your email address will not be published.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