Synthetic Human Avatar Images for Cervical Posture Analysis, 16,000 Renders

16,000 synthetically generated frontal-view human avatar images with ground-truth labels for lateral-shift estimation. The dataset was created by Roland Stenger using a scripted Blender pipeline with the MPFB add-on and was last updated on May 31, 2026. It includes raw renders, JSON labels with precise rig parameters, preprocessed segmentation maps, and trained model weights.

Use Cases

• Train a regression model for lateral-shift estimation based on synthetic avatar images.
• Develop segmentation models for human anatomy using the provided preprocessed segmentation maps.
• Study the impact of pose parameters like yaw, roll, and pitch on cervical posture prediction.
• Benchmark synthetic data performance against real-world medical imaging datasets for posture analysis.

Strengths

• Contains 16,000 synthetic images, providing a substantial volume for model training.
• Ground-truth labels are derived directly from the avatar rig, providing continuous values without human rating noise.
• Includes multiple data modalities: raw RGB images, JSON labels, segmentation arrays, and trained model weights.
• Fully synthetic nature eliminates privacy concerns associated with real human data.

Limitations

• The dataset size is listed as 16.0 B, indicating a very limited scope for the actual downloadable package.
• Row count for the primary image data is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics for JSON labels must be inferred from the provided example.

Provenance

Source

figshare, author Roland Stenger.

Collection Method

Synthetically generated via a scripted Blender pipeline using the MPFB add-on for human model generation.

Freshness

Last updated 2026-05-31 14:59:42; freshness should be verified.