The underground station where images are captured

Sample images in the dataset

Download

Download
Including raw images, features, and train/test partitions.

Details

The QMUL underGround Re-IDentification (GRID) dataset contains 250 pedestrian image pairs. Each pair contains two images of the same individual seen from different camera views. All images are captured from 8 disjoint camera views installed in a busy underground station. The figures beside show a snapshot of each of the camera views of the station and sample images in the dataset. The dataset is challenging due to variations of pose, colours, lighting changes; as well as poor image quality caused by low spatial resolution.

There are two folders:
Folder 'probe' contains 250 probe images captured in one view.
Folder 'gallery' contains 250 true match images of the probes captured in other views. Besides, there are a total of 775 additional images that do not belong to any of the probes. These extra images should be treated as a fixed portion in the testing set during cross validation.

The dataset is intended for research purposes only and as such cannot be used commercially. Please cite the following publication(s) when this dataset is used in any academic and research reports.

References

  1. On-the-fly Feature Importance Mining for Person Re-Identification
    C. Liu, S. Gong, and C. C. Loy
    Pattern Recognition, vol. 47, no. 4, pp. 1602-1615, 2014 (PR)
    DOI PDF Project Page
  2. Person Re-Identification
    S. Gong, M. Cristani, S. Yan, C. C. Loy (Eds.)
    Springer, January 2014
    DOI Preface Introduction
  3. Evaluating Feature Importance for Re-Identification
    C. Liu, S. Gong, C. C. Loy, and X. Lin
    In Gong, Cristani, Yan, Loy (Eds.), Person Re-Identification, Springer, January 2014
    PDF
  4. Person Re-Identification by Manifold Ranking
    C. C. Loy, C. Liu, and S. Gong
    IEEE International Conference on Image Processing, pp. 3567-3571, 2013 (ICIP)

    PDF Poster Project Page | Codes
  5. Time-Delayed Correlation Analysis for Multi-Camera Activity Understanding
    C. C. Loy, T. Xiang, and S. Gong
    International Journal of Computer Vision, vol. 90(1), pp. 106-129, October 2010 (IJCV)
    DOI PDF Project Page | Codes
  6. Multi-Camera Activity Correlation Analysis
    C. C. Loy, T. Xiang, and S. Gong
    IEEE Conference on Computer Vision and Pattern Recognition, pp. 1988-1995, 2009 (CVPR)
    * Oral presentation
    PDF Project Page | Codes

Benchmarking Results

The matching performance given in the table below was measured using the averaged cumulative match characteristic (CMC) curve over 10 trials. For feature extraction please refer to this paper for more details. The number of train/test images were set to (125 paired images)/(125 paired images + 775 non-paired images).

Below are the methods that perform benchmarking on GRID:

  1. RankSVM - B. Prosser, W. Zheng, S. Gong and T. Xiang. Person Re-Identification by Support Vector Ranking. In Proc. British Machine Vision Conference (BMVC), Aberystwyth, Wales, September 2010.
  2. PRDC - W. Zheng, S. Gong and T. Xiang. Re-identification by Relative Distance Comparison. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), June 2012.
  3. MRank-RankSVM - C. C. Loy, C. Liu, and S. Gong. Person Re-identification by Manifold Ranking. IEEE International Conference on Image Processing (ICIP), 2013.
  4. MRank-PRDC - C. C. Loy, C. Liu, and S. Gong. Person Re-identification by Manifold Ranking. IEEE International Conference on Image Processing (ICIP), 2013.
  5. MtMCML - L. Ma, X. Yang, D. Tao. Person Re-Identification Over Camera Networks Using Multi-Task Distance Metric Learning. IEEE Transaction on Image Processing, vol. 23, no. 8, pp. 3656-3670, 2014.
  6. LCRML - J. Chen, Z. Zhang and Y. Wang. Relevance Metric Learning for Person Re-identification by Exploiting Global Similarities. IEEE International Conference on Pattern Recognition (ICPR), pp. 1657-1662, 2014.
  7. XQDA - S. Liao, Y. Hu, X. Zhu, S. Z. Li. Person Re-identification by Local Maximal Occurrence Representation and Metric Learning. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
  8. PolyMap - D. Chen, Z. Yuan, G. Hua, N. Zheng, J. Wang. Similarity Learning on an Explicit Polynomial Kernel Feature Map for Person Re-Identification. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
  9. KEPLER - N. Martinel, C. Micheloni, G. L. Foresti. Kernelized Saliency-Based Person Re-Identification Through Multiple Metric Learning. IEEE Transaction on Image Processing, vol. 24, no. 12, pp. 5645-5658, 2015.
  10. NLML - S. Huang, J. Lu, J. Zhou, A. K. Jain. Nonlinear Local Metric Learning for Person Re-identification. arXiv:1511.05169v1, 2015.
  11. SSDAL + XQDA - Chi Su, Shiliang Zhang, Junliang Xing, Wen Gao, and Q, Tian. Deep Attributes Driven Multi-Camera Person Re-identification. arXiv:1605.03259, 2016
  12. DR-KISS - D. Tao, Y. Guo, M. Song, Y. Li, Z. Yu, and Y. Y. Tang. Person Re-Identification by Dual-Regularized KISS Metric Learning. IEEE Transaction on Image Processing, vol. 25, no. 6, pp. 2726-2738, 2016
r=1r=5r=10r=15r=20
L1-norm4.4011.6816.2419.1224.80
PRDC9.68 22.00 32.96 38.96 44.32
RankSVM10.24 24.56 33.28 39.44 43.68
XQDA10.48 28.08 38.64 46.32 52.56
LCRML10.68 25.76 35.04 42.08 46.48
MRank-PRDC11.12 26.08 35.76 41.76 46.56
MRank-RankSVM12.2427.8436.3242.2446.56
MtMCML14.08 34.64 45.84 52.88 59.84
PolyMap16.30 35.80 46.00 52.80 57.60
XQDA + LOMO 16.56 33.84 41.84 47.68 52.40
KEPLER 18.40 39.12 50.24 57.04 61.44
NLML 24.54 35.86 43.53 - 55.25
SSDAL + XQDA 22.40 39.20 48.00 -- 58.40
DR-KISS 20.60 39.30 51.40 - 62.60
Example results from PRDC
Success cases
Failure cases

Acknowledgements

We would like to thank the UK MOD who have made the video footage available to the Queen Mary University of London. Additionally we thank David Russell for processing the video data.

Other Re-Identification Datasets