OpenThermalPose: Open-Source Annotated Thermal Human Pose Datasets

This repository contains open-source dataset for thermal human pose estimation: OpenThermalPose. Dataset provide annotations of human poses in thermal imagery, suitable for training and evaluating pose estimation models. The dataset has extension OpenThermalPose2

OpenThermalPose consists of 6,090 images depicting 31 subjects and 14,315 annotated human instances. Each instance includes a bounding box and 17 annotated keypoints, following the MS COCO Keypoint dataset convention. The dataset captures various scenarios, including fitness exercises, multi-person activities, and outdoor walking under diverse weather conditions. as individuals sitting indoors.

Dataset Statistics

Dataset	Images	Subjects	Instances	Keypoints	Activities	Environments
OpenThermalPose	6,090	31	14,315	17	Fitness exercises, multi-person activities, walking	Indoor, Outdoor
OpenThermalPose2	11,391	170	21,125	17	Fitness exercises, multi-person activities, walking, sitting	Indoor, Outdoor

Citations

OpenThermalPose:

@INPROCEEDINGS{10581992,
  keywords={Privacy;Annotations;Source coding;Pose estimation;Lighting;Medical services;Motion capture},
  doi={10.1109/FG59268.2024.10581992}}