Data for ‘Automatic Pose Estimation in Newborn Infants: Lessons from the Baby Grow Study’

This data is related to the paper "<b>Automatic Pose Estimation in Newborn Infants: Lessons from the Baby Grow Study</b>".It includes accuracy (OKS), speed and efficiency data for each HPE model in different video conditions.Video conditions:Age: 2, 4 and 8 weeksBackground: Grey, Black and ColourfulClothing: Common, BG and VestRecording angle: Top, Front and BottomLighting: With and without ShadowAll conditions<b>Below is the list of items you can see in this repository:</b><b> </b>|--- <b>Models detections and Ground-truth_Automatic_Pose_Estimation_Baby_Grow</b><b>CVAT_Ground-truth.zip</b>This file includes all Ground-truth (GT) detections for each video as JPG files, showing the location of each keypoint on the body, and a CSV file showing the numerical location of each landmark.<b>MediaPipe.zip</b>This file includes all MediaPipe detections for each video as JPGs, showing the location of each keypoint on the body relative to the GT, and a CSV file showing the numerical locations of each MediaPipe landmark detection.<b>OpenPose.zip</b>This file includes all OpenPose detections for each video as JPGs, showing the location of each keypoint on the body relative to the GT, and a CSV file showing the numerical locations of each OpenPose landmark detection.<b>PCT.zip</b>This file includes all PCT detections for each video as JPGs, showing the location of each keypoint on the body relative to the GT, and a CSV file showing the numerical locations of each PCT landmark detection.<b>RTMpose.zip</b>This file includes all RTMpose detections for each video as JPGs, showing the location of each keypoint on the body relative to the GT, and a CSV file showing the numerical locations of each RTMpose landmark detection.<b>Sapiens.zip</b>This file includes all Sapiens detections for each video as JPGs, showing the location of each keypoint on the body relative to the GT, and a CSV file showing the numerical locations of each Sapiens landmark detection.<b>ViTPose.zip</b>This file includes all ViTPose detections for each video as JPGs, showing the location of each keypoint on the body relative to the GT, and a CSV file showing the numerical locations of each ViTPose landmark detection.|--- <b>Reliability_Automatic_Pose_Estimation_Baby_Grow.zip</b>This file includes all the final data used for the statistical analysis of each variable, including APS, OKS, PCK, Speed, Efficiency and missing and extra detections by models.|--- <b>Codes_Automatic_Pose_Estimation_Baby_Grow.zip</b>This file includes all Python code for the calculation of OKS, PCK, Efficiency, FP and ICC analysis.|--- <b>Reliability_Automatic_Pose_Estimation_Baby_Grow.zip </b>This file includes all the Ground-truth and the Coder_1 and Coder_2 detection used for the reliability check.|--- <b>201123 Baby Grow Parent Instructions Video only with exercises V3.pdf</b>This is the main instruction sent to the participants for video recordings.Below is the abstract of the paper:<b>Abstract</b><br>Advances in computational techniques -particularly machine learning- have expanded opportunities to analyse early infant motor repertoires, especially in naturalistic settings. The aim of this study was to evaluate the strengths, limitations, and performance of state-of-the-art pose estimation algorithms in challenging, home-based video conditions. We analysed 22 videos recorded by parents using mobile phones from eight newborns in the Baby Grow study, at 2, 4, and 8 weeks of age. The videos varied in clothing (common onesie, Baby Grow, vest), background (grey, black, coloured), lighting (with/without shadows), and camera angles (top, front, bottom). From these, 2,640 frames were extracted and manually annotated to serve as ground truth.We tested demo versions of MediaPipe, OpenPose, PCT, RTMpose, Sapiens, and VitPose, and evaluated performance using Object Keypoint Similarity (OKS), Percentage of Correct Keypoints (PCKh), speed and accuracy. RTMpose showed the highest overall accuracy, while MediaPipe had the fastest processing speed. However, when balancing speed and accuracy at ratios of 70:30, 50:50, and 30:70, MediaPipe’s speed compensated for its lower accuracy, making it a strong candidate for practical applications. Model performance varied under different environmental conditions, with RTMpose, Sapiens, and VitPose being the most robust.As infant movement research increasingly shifts to real-world environments, selecting appropriate models and ensuring video quality are essential. Our findings highlight: (1) new models outperform legacy tools like OpenPose, and (2) video context and model selection significantly affect pose estimation accuracy.