Nataraj Jammalamadaka, Andrew Zisserman, Marcin Eichner, Vittorio Ferrari and C. V. Jawahar
Overview
This dataset is derived from Hollywood movies and “Buffy, the vampire slayer” series. The images are randomly sampled from the aforementioned videos. All humans with more than three parts visible are annotated with upper body stickmen (6 parts: head, torso, upper and lower arms).
More details
Annotations from Hollywood movies: A total of about 6000 frames are sampled from ten Hollywood movies viz., “About a Boy”, “Apollo 13”, “Four Weddings and a Funeral”, “ Forrest Gump”, “ Notting Hill”, “Witness”, “Gandhi”, “Love Actually”, “The Graduate” and “Groundhog day”. It has 11,639 ground truth annotations. These annotations offers a large variety of poses, people and backgrounds.
Annotations from the TV show "Buffy the vampire slayer": A total of 499 frames are sampled from episodes 2 and 3 of season 5 of this TV show. It contains a total of 755 annotations. This dataset is mainly introduced to add to the numbers of Buffy stickmen dataset.
Downloads
Please cite [1] if you use any of the above datasets. For the procedure to benchmark a human pose estimation algorithm, please refer to Buffy stickmen webpage.
A few more Examples
Publications
Acknowledgements
We are grateful for financial support from the UKIERI, the ERC grant VisRec no. 228180 and the Swiss National Science Foundation. We also thank Rajan and Nandini, the IIIT annotation team, for annotating the dataset.