Let's play

Four Stages of Dataset Grief

Download the paper

A. Torralba, A. Efros. Unbiased Look at Dataset Bias. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2011. (pdf)

Related work

In the paper, we lamented the fact that there is virtually no effort being made in papers to systematically compare algorithm performance "across" datasets. Since the paper came out, we have learned of a few heroic efforts in this direction. The following is a list of published works in recognition which report cross-dataset generalization performance:

Please e-mail the authors if you feel your paper should be in this list. All citations will be included in our upcoming journal submission.

News

The "Name That Dataset" game was played during the CVPR'2011 poster session. The winner, with 100% correct answers, was Prof. Lana Lazebnik, who will be getting a fancy dinner from us one of these days.

Accessibility

Acknowledgments

The authors would like to thank the Eyjafjallajokull volcano as well as the wonderful kirs at the Buvette in Jardin du Luxembourg for the motivation (former) and the inspiration (later) to write this paper. This work is part of a larger effort, joint with David Forsyth and Jay Yagnik, on understanding the benefits and pitfalls of using large data in vision. The paper was co-sponsored by ONR MURIs N000141010933 and N000141010934. No graduate students were harmed in the production of this paper. Authors are listed in order of increasing procrastination ability.