innovarul / sysu_mm01_pythoneval

(Re-)implementation of evaluation code for SYSU_MM01 dataset in python

SYSU_MM01_pythoneval

(Re-)implementation of evaluation code for SYSU_MM01 dataset in python

The code is a python translation of the Matlab-code provided by SYSU-MM01 dataset authors here.

Dataset download

Baiduyun: http://pan.baidu.com/s/1gfIlcmZ

Dropbox: https://www.dropbox.com/sh/v036mg1q4yg7awb/AABhxU-FJ4X2oyq7-Ts6bgD0a?dl=0

Please refer the readme file of original Matlab version of evaluation code for more information about the dataset and evaluation protocol.

Usage

As per original protocol, the feature vectors are to be stored in .mat files for each camera in a particular directory. Then you can run the evaluation to get the performance numbers.

check the file src/evaluate_SYSU_MM01.py to locate the function call for evaluation.

evaluate_results(feature_dir, prefix, mode, number_shot, total_runs=10)

# feature_dir = the directory path where the camera based features are stored
# prefix = prefix of the files or model name
# mode = all | indoor
# number_shot = 1 = single-shot | 10 = multi-shot
# total_runs = number of test iterations to run (usually 10, as given in the original paper)

Citation

You can cite the original paper from Wu et al., if you use the dataset.

Ancong Wu, Wei-Shi Zheng, Hong-Xing Yu, Shaogang Gong and Jianhuang Lai. RGB-Infrared Cross-Modality Person Re-Identification. IEEE International Conference on Computer Vision (ICCV), 2017.

sysu_mm01_pythoneval's Issues

Usage of rand_perm_cam.mat file

Hi, Thanks a lot for open souring this implementation!
I have a doubt regarding the contents of rand_perm_cam.mat file. I understand it has some fixed split of gallery images. But can you please elaborate why is it needed for the evaluation of map and cmc scores because in some other others for calculating rb ir cross modal person reid accuracy, it is not getting used.
Thanks