Data Sets
The following two datasets were published in:
S. Bak, E. Corvee, F. Bremond, M. Thonnat,
"Boosted Human Re-identification using Riemannian Manifolds", Image and Vision Computing, Special Issue on Manifolds for Computer Vision, 2011.
iLIDS-MA
This dataset contains 40 individuals extracted from two cameras. For each individual
46 frames have been annotated manually from both cameras. Therefore we have
40 x 2 x 46 = 3680 annotated images.
The sample images from iLIDS-MA dataset. Top and bottom lines
correspond to images from different cameras. Columns illustrate the same person.
iLIDS-AA
The manually annotated dataset (i-LIDS-MA) does not reflect a real video surveillance
scenario where humans are detected and tracked automatically. Consequently, we
have applied HOG-based human detector and tracker to obtain multiple images of
individuals seen from both cameras. In this case, detection and tracking results
are noisy which makes the dataset more challenging. This dataset contains 100
individuals. For each individual we have extracted automatically a different number
of frames depending on tracking difficulties.
The sample images from iLIDS-AA dataset. Top and bottom lines
correspond to images from different cameras. Columns illustrate the same person.
How to get iLIDS-MA/iLIDS-AA?
For downloading this data sets, you need to have the right to access the i-LIDS Multiple-Camera Tracking Scenario (MCTS). Please contact Slawomir.Bak[at]inria.fr for more information.