Q: After I download the MediaMill Challenge,I find the MD5 checksum of the downloaded file is different from what you provide. When I unzip the file, some errors are shown like "unexpected end of archive, read error in the file, the file is corrupt". What is wrong?
A: Most browsers can not handle files larger than 2GB. To prevent that the size of the tarball is misinterpreted, you have to use the wget --ignore-length option. See also the detailed download instruction.
Q: We are working on the MediaMill Challenge and have some difficulties in reproducing your results. Can it be caused by the fact that we do not use the same version of LIBSVM?
A: It can make a difference. Georges Quénot (Laboratoire d'Informatique de Grenoble) reported that when using LIBSVM 2.85 there are slight variations in the number of support vectors and in displayed numbers (#iter, rho, ..., about a few per thousand). Any derived figure, including APs are also slightly different, generally within a percent on relative values.

Address your questions to:

Readme First
Cees G.M. Snoek, Marcel Worring, Jan C. van Gemert, Jan-Mark Geusebroek, and Arnold W.M. Smeulders. The Challenge Problem for Automated Detection of 101 Semantic Concepts in Multimedia. In Proceedings of ACM Multimedia, pp. 421-430, Santa Barbara, USA, October 2006.