PoseFES

News

• 2020-04-27 - PoseFES dataset released with images, bounding boxes and keypoints.
• 2023-04-05 - Paper accepted by CVPR OmniCV 2023 workshop.

Abstract

Please download and sign the dataset release agreement, and send it to Jingrui Yu for access of the dataset. The dataset is for non-commercial usage only.

We created the real-world dataset PoseFES for evaluation purposes by extending the FES Dataset with one scenario and pose annotations. It consists of two sequences, which have been recorded in a laboratory apartment with an omnidirectional fisheye camera. The image resolution is 1680 × 1680 pixels. The first sequence, Scenario 1 (Sc1), contains 400 frames (Record 00000.png – Record 00399.png), in which three persons walk through the apartment performing daily activities. Overlapping of persons is kept very seldom for this sequence. Scenario 2 (Sc2) contains 301 frames (Record 00600.png – Record 00900.png), in which a maximum of eight persons appear at the same time. Heavy overlapping is present in most frames of this sequence. There are 735 and 2161 instances in Sc1 and Sc2, respectively.

Axis-aligned bounding boxes and keypoints are annotated for the dataset. The bounding boxes are generated using OmniPD and then adjusted manually. 17 keypoints are annotated for all persons, which conform to the keypoints provided by COCO. Two extra keypoints, shoulder center and hip center are extrapolated by averaging shoulder and hip keypoints, respectively. Annotations are available in CVAT and COCO format.

Paper

• CVPR 2023 workshop paper
• arXiv

BibTeX

If you use the data set in your work, please don't forget to cite:

@InProceedings{Yu_2023_CVPR,
    author    = {Yu, Jingrui and Scheck, Tobias and Seidel, Roman and Adya, Yukti and Nandi, Dipankar and Hirtz, Gangolf},
    title     = {Human Pose Estimation in Monocular Omnidirectional Top-View Images},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2023},
    pages     = {6410-6419}
}