State of the art on the Market-1501+500k dataset

In this page, will summarize the state-of-the-art methods on Market-1501+500k dataset. We report both mAP and rank-1 accuracy under different gallery sizes. Note that this may not be the only performance measurement. Other metrics, such as retrieval time, are also important. Should you have any inquiries please let me know at

Reference Gallery size Notes
"Scalable person re-identification: a benchmark", Liang Zheng, Liyue Shen, Lu Tian, Shengjin Wang, Jingdong Wang, Qi Tian, ICCV 2015 mAP13.9411.4410.528.66BoW, Euclidean distance, single query
mAP13.8510.889.757.56BoW+ANN [1], single query
mAP18.3815.9514.8812.60BoW, Euclidean distance, multiple query
mAP18.2615.0913.7510.92BoW+ANN [1], multiple query
"Person re-identification: Past, Present and Future", Liang Zheng, Yi Yang, Alexander Hauptmann, Arxiv 2016 rank-173.6972.1571.5570.67 ResNet50 baseline. The 2,048-dim feature from pool5 is used under Euclidean distance. Code can be accessed here.
Current state of the art
"A Discriminatively Learned CNN Embedding for Person Re-identification", Zhedong Zheng, Liang Zheng and Yi Yang, Arxiv 2017. rank-1 79.51 73.7871.5068.26A two-stream network based on ResNet50, single query. Code is available upon request.
mAP 59.87 52.2849.1145.24
"Improving Person Re-identification by Attribute and Identity Learning", Yutian Lin, Liang Zheng, Zhedong Zheng, Yu Wu and Yi Yang, Arxiv 2017. rank-1 83.99 79.8978.2075.44Attribute and ID classification is jointly learned. ResNet50 is used as backbone. Pool5 feature is used under Euclidean distance, single query. Attribute labels can be accessed here.
mAP 62.83 56.4653.5849.78
"In Defense of the Triplet Loss for Person Re-Identification", Alexander Hermans, Lucas Beyer and Bastian Leibe, Arxiv 2017. rank-1 84.92 79.6977.8874.70single query. The triplet-loss based network is fine-tuned. Image size: 256x128. The last layer in ResNet is replaced with one 1,024-dim layer and one 128-dim layer. Batch normalization.
mAP 69.14 61.9358.7453.63


[1] J. Wang and S. Li. Query-driven iterated neighborhood graph search for large scale indexing. In ACM MM, 2012.