The SYSU-30k dataset contains 30,000 pedestrian identity categories, which is approximately the same as the CUHK03 dataset and Market-1501 datasetIf a pedestrian identity category is equivalent to an object category, then SYSU-30k is equivalent to ImageNet dataset30 times of that. The dataset contains a total of 29,606,918 images. The SYSU-30k dataset not only provides an evaluation platform for the weakly supervised ReID problem, but also a test set that meets the challenges of real-world scenarios.

Citation

@inproceedings{Wang2020Weakly_tnnls, title={Weakly Supervised Person Re-ID: Differentiable Graphical Learning and A New Benchmark}, author={Guangrun Wang and Guangcong Wang and Xujie Zhang and Jianhuang Lai and Zhengtao Yu and Liang Lin}, booktitle={IEEE Transactions on Neural Networks and Learning Systems (T-NNLS)}, year={2020} }