Rick Szeliski (Microsoft Research), Ramin Zabih (Cornell University), Daniel Scharstein (Middlebury College),
Olga Veksler (University of Western Ontario), Vladimir Kolmogorov (University College London),
Aseem Agarwala (University of Washington), Marshall Tappen (MIT), and Carsten Rother (Microsoft Research).

This site contains the code and results (images and plots) accompanying our paper

A Comparative Study of Energy Minimization Methods for Markov Random Fields
R. Szeliski, R. Zabih, D. Scharstein, O. Veksler, V. Kolmogorov, A. Agarwala, M. Tappen, and C. Rother.
In Ninth European Conference on Computer Vision (ECCV 2006), volume 2, pages 19-26, Graz, Austria, May 2006.

Code

Note: The current version contains ICM, Graph cuts swap and expansion moves, and max-product belief propagation. It does not yet contain TRW / TRW-S, which we hope to add in a future version. We will also add more sample applications that use the library.

Results

Stereo matching problems:

• Tsukuba1 (384 x 288, 16 labels)
• Venus1 (434 x 383, 20 labels)
• Teddy1 (450 x 375, 60 labels)

Photomontage problems:

• Panorama (1071 x 480, 7 labels)
• Family (752 x 566, 5 labels)

Binary image segmentation problems:

• Flower (600 x 450, 2 labels)
• Sponge (640 x 480, 2 labels)
• Person (600 x 450, 2 labels)

Image restoration and inpainting problem:

• Penguin (122 x 179, 256 labels)