ChokePoint Dataset Bounding Box Annotations

Bounding box annotations for the G1 and G2 sets of the ChokePoint dataset, provided as a supplementary material to:

S. Alver and U. Halici, "Attentive Deep Regression Networks for Real-Time Visual Face Tracking in Video Surveillance," Submitted, 2019.

Description

This repository contains bounding box annotations for the G1 and G2 sets (sets for the baseline verification protocol) of the ChokePoint dataset. For the ease of training (or development) and evaluation, we provide two folders: G1 and G2. These folders have a train_annotation.txt file that contains the training annotations for that folder. They also have 216 separate eval_annotation_seq_X.txt files that contain the evaluation annotations for the 216 different sequences in the evaluation set. We provide them separately so that the performance on each of the evaluation sequences can be examined individually. Each of the lines in these files are in the form of: file_directory+file_name, top_left_row, top_left_column, width, height.

NOTE: The original dataset only has person ID and eye location annotations, which makes it incompatible with the task of visual face tracking.

NEW: Also added the bounding box results of the trackers described in the paper so that comparisons can be done. There are two kinds of results: 1) Results where the trackers are not reinitialized after target loss (in the not_reinit folders) 2) Results where the trackers are reinitialized after complete target loss (in the reinit folders). It should be noted that these results do not contain the bounding boxes for the first frames in the sequences as they are only used for initializing the trackers.

Citation

If you find this work to be useful for your studies, please cite (using the BibTeX entries) the following two articles:

@misc{alver_2019,
  Author = {Safa Alver and Ugur Halici},
  Title = {Attentive Deep Regression Networks for Real-Time Visual Face Tracking
  in Video Surveillance},
  Year = {2019},
  Eprint = {arXiv:1908.03812},
}

@inproceedings{wong_cvprw_2011,
   Author = {Yongkang Wong and Shaokang Chen and Sandra Mau and Conrad Sanderson
   and Brian C. Lovell},
   Title = {Patch-based Probabilistic Image Quality Assessment for Face Selection
   and Improved Video-based Face Recognition},
   Booktitle = {IEEE Biometrics Workshop, Computer Vision and Pattern Recognition
   (CVPR) Workshops},
   Year = {2011},
   Pages = {81-88},
   Month = {June},
   Publisher = {IEEE}
}