This dataset is for research purpose only and is introduced in the following paper,

P. Liang, Y. Wu, H. Lu, J. Xu, Y. Wang and H. Ling. "Planar Object Tracking in the Wild: A Benchmark", IEEE International Conference on Robotics and Automation (ICRA), 2017, under review 

Each video sequence has 501 frames, the first frame is used for initialization. For each sequence, we annotate every other frame, and the annotation contains the following three files,

* [name]_gt_points.txt contains the ground truth positions of the four reference points. The format is [point_1_x point_1_y point_2_x point_2_y point_3_x point_3_y point_4_x point_4_y].
* [name]_flag.txt indicates the status of each frame. (0:normal; 1:occluded for about more than half; 2: out of view for about more than half; 3: heavily blurred.)
* [name]_gt_homography.txt contains ground truth homography transformation (3x3 matrix) that projects the initial four reference points to a given frame.

If you have any questions, please contact Pengpeng Liang at pliang AT temple.edu or Haibin Ling at hbling AT temple.edu.