I send here donated features about face detection. I followed as much as possible the format defined in: > I have left the shot timestamps inside and for each shot there is a Keyword line regardless whether there is on not faces and the number of faces is included is the keyword as: 0 face 1 face 2 faces .... This face extraction has been achieved as a cooperation between CLIPS, LIT (Laboratories of Information Technologies, Singapore, http://w= ww.lit.a-star.edu.sg/) and, indirectly, CMU. The data was obtained as follows: CLIPS extracted one keyframe per shot (as the one with the highest contrast) and for each keyframe, LIT run its version of a face detection freeware available from CMU (http://www-2.cs.cmu.edu/~tom/faces.html, I think). We made some checking between the detector output and the keyframes and it turned out that the result was quite good even if the faces are not so frontal. It seems that there is much less false positive than false negative (even for frontal faces). This means that there is in general actually more faces than recovered. Also, this is for keyframes only and the actual number of faces can change during the shot. No data of the TREC video 2002 was used at all for training any part of the system used for the extraction. Therefore, this is a "type B" donation. I have left "cmu" in the name of the tar file and directory since their detector has made a significant part of the work but I didn't check with them whether they wish to be explicitely associated to this donation or not. I just guessed yes. We did not succeed in producing other features with a reasonable quality in time. Georges. Email: Georges.Quenot@imag.fr CLIPS-IMAG, 385, rue de la Biblioth=E8que, B.P. 53, 38041 Grenoble Cedex = 9 Tel: (33-4) 76 63 58 55, Fax: (33-4) 76 44 66 75