Baby Grow Study: Model Performance Data for Infant Pose Estimation Across Video Conditions

Saber Sotoodeh's data for the paper 'Automatic Pose Estimation in Newborn Infants: Lessons from the Baby Grow Study'. It includes accuracy, speed, and efficiency metrics for six human pose estimation models across varied video conditions. The dataset was last updated on April 30, 2026, and is licensed under CC-BY-4.0.

Use Cases

• Benchmarking pose estimation models (MediaPipe, OpenPose, PCT, RTMpose, Sapiens, ViTPose) based on OKS, PCK, speed, and efficiency metrics.
• Analyzing the impact of video conditions (infant age, background, clothing, recording angle, lighting) on model performance.
• Training or fine-tuning models for infant pose detection using provided ground-truth keypoint annotations.
• Studying model reliability by examining patterns in missing and extra detections across different scenarios.

Strengths

• Includes performance data for six distinct human pose estimation models.
• Evaluates models across multiple controlled video conditions: 3 infant ages, 3 backgrounds, 3 clothing types, 3 recording angles, and 2 lighting states.
• Dataset size is 15.0 GB, containing both numerical CSV data and visual JPG comparisons.
• Provides ground-truth annotations for validation, stored as JPG files and CSV coordinates.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific study setup and conditions captured.

Provenance

Source

figshare, author Saber Sotoodeh.

Collection Method

Likely collected from video recordings of newborn infants under controlled conditions for the Baby Grow Study.

Time Range

null

Freshness

Last updated 2026-04-30 14:24:08; freshness should be verified.

Geography

null