(c) Copyright 2015-2016 PRA Lab - Pattern Recognitions and Applications Lab, University of Cagliari. Bahram Lavi lavi.bahram@diee.unica.it
- Introduction:
This code provides the implementation of the multi-stage ranking system from our paper 'A Multi-Stage Ranking Approach for Fast Person Re-Identification'. This code implements only the multi-stage system and not features extraction and neither obtaining the simplified versions of any given descriptor. In the 'data/' directory we provide also, as example, the pre-computed matching scores for SDALF descriptor on VIPeR, i-LIDS, and ETHZ1 data sets.
If one interested in use of different descriptors(e.g. gBiCov, LOMO, and MCM), the matching scores might be already computed for all the available templates in the gallery set for each simplified version, and stated the MAT file in the 'data/' directory. Then, the corresponding 'MAT' file name might be specified in 'Load_Scores.m' to be loaded for use of the multi-stage system.
- Running the code:
To run the code you might run 'MainMS.m' at this directory. You may simply initialize the parameters by yourself such as: the descriptor, data set, and number stages. This code only implemented for two- and three-stage systems so far.
- CMC curves in our paper:
Also, the figures presented in our paper can be found in 'CMC_cuves/' directory.
- Troubleshooting:
If you have any problem or find a bug mail lavi.bahram@diee.unica.it.