The goal of this website is to provide tools to evaluate and compare models and methods for learning to recognize basic entities in images in an unsupervised setting. In other words, we want to discover the objects present in the images by analyzing unlabeled data and searching for re-occurring patterns.

This work is described in detail in:

T. Tuytelaars, C.H. Lampert, M. B. Blaschko, W. Buntine, "Unsupervised Object Discovery: A comparison", International Journal on Computer Vision, doi:10.1007/s11263-009-0271-8, 2009. pdf

We follow the evaluation protocol as proposed by Sivic et al., constructing a dataset composed of images from a fixed number of predefined categories. In particular, we use subsets of the Caltech256 dataset as well as the MSRC2 dataset. We propose different evaluation metrics, based on conditional entropy, purity, or F1-measure. Scripts for performing these evaluations are provided, as well as precomputed image representations based on local features. Within this framework, we have compared several baseline methods, methods based on latent variable models, as well as spectral clustering methods. Code for some of these is made available, and the most important results are summarized.