Gamified VR Human Posing Application for Deep Learning Synthetic Data Generation

As deep learning systems increasingly depend on human image and video data, privacy concerns make real data collection costly and time-consuming. Many companies turn to synthetic data, which is easier to acquire, GDPR-compliant, and helps reduce model bias. However, creating diverse human poses remains labor-intensive. To address this, we introduce VRPoser, a VR-based serious game and tool for intuitive pose creation. Users take on the role of artists, completing story-driven quests to pose mannequins. These poses are converted to OpenPose skeletons and used with ControlNet and Stable Diffusion to generate synthetic images. When evaluated for usability, VRPoser was found to be intuitive, fun, and efficient. Comparisons with standard posing software show that VRPoser achieves similar results in less time.