Skip to content

liminxian/DukeMTMC-SI-Tracklet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 

Repository files navigation

DukeMTMC-SI-Tracklet

DukeMTMC-SI-Tracklet [1] is an auto-detected tracklets subset of the DukeMTMC tracking dataset [2] for tracklet-based person re-identification.

We built this new tracklet person re-id dataset as follows. We first deployed an efficient deep learning tracker that leverages a COCO+PASCAL trained SSD for pedestrian detection and an ImageNet trained Inception for person appearance matching. Applying this tracker to all DukeMTMC raw videos, we generated 19,135 person tracklets. Due to the inevitable detection and tracking errors caused by background clutters and visual ambiguity, these tracklets may present typical mistakes (e.g.ID switch) and corruptions (e.g. occlusion).

We finally obtained 12,647 person tracklets from 1,788 unique IDs. To match DukeMTMC-ReID and DukeMTMC-VideoReID, we set the same 702 training IDs with the remaining 1,086 people for performance test (missing 14 test IDs against DukeMTMC-ReID due to tracking failures).

Dataset Offline

Due to unforeseen circumstances, the DukeMTMC benchmark dataset and its derivatives including the DukeMTMC-SI-Tracklet dataset have been requested to be removed from online until further notice. We sincerely apologize for any inconvenience caused.

References

Please cite the following two papers if this dataset helps your research.

@article{LiTPAMI2019,
 title={Unsupervised Tracklet Person Re-Identification},
 author = {Li, Minxian and Zhu, Xiatian and Gong, Shaogang},
 journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
 year={2019}
}

@inproceedings{ristani2016MTMC,
  title = {Performance Measures and a Data Set for Multi-Target, Multi-Camera Tracking},
  author = {Ristani, Ergys and Solera, Francesco and Zou, Roger and Cucchiara, Rita and Tomasi, Carlo},
  booktitle = {European Conference on Computer Vision workshop on Benchmarking Multi-Target Tracking},
  year = {2016}
}

About

Unsupervised Tracklet Person Re-Identification

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published