Efficient PSD Constrained Asymmetric Metric Learning for Person Re-identification

Download  |   Results  |   Contact  |   References

Problems with existing metric learning methods:

  • Applying PSD: expensive
  • No PSD: noisy
  • pos/neg samples: largely unbalanced


  • APG solution to the PSD constrained logistic metric learning problem
  • Asymmetric pos/neg sample weights to balance pos/neg costs


  • PSD+APG leads to low rank and smooth metric
  • APG solution is fast in convergence
  • PSD and asymmetric weights lead to notable improvements




Fig. 1. Fast rank shrinkage.

Fig. 2. Effect of low rank selection.



Fig. 3. Improvement by PSD.

Fig. 4. Improvement by weighting.

Table 1. Summary of results (%) for the proposed MLAPG algorithm

Database Rank 1 Rank 5 Rank 10 Rank 15 Rank 20
VIPeR 40.73 69.94 82.34 88.48 92.37
QMUL Grid 16.64 33.12 41.20 48.16 52.96
CUHK01 64.24 85.41 90.84 93.35 94.92
CUHK03 Labeled 57.96 87.09 94.74 97.04 98.00
CUHK03 Detected 51.15 83.55 92.05 95.30 96.90

Note: CMC curves of the proposed MLAPG algorithm can be downloaded in cmc_curves.zip



National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences.


[1]Shengcai Liao and Stan Z. Li, “Efficient PSD Constrained Asymmetric Metric Learning for Person Re-identification.” In IEEE International Conference on Computer Vision (ICCV 2015), December 11-18, Santiago, Chile, 2015. [pdf][poster]

Last updated: Dec. 8, 2015



Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s