Related Experiment Video
Updated: Jan 11, 2026
07:25
An Objective and Child-friendly Assessment of Arm Function by Using a 3-D Sensor
Published on: February 12, 2018
7.2K
Quality Assessment of 3D Human Animation: Subjective and Objective Evaluation
IEEE Transactions on Visualization and Computer Graphics
|November 12, 2025
Summary
We developed a new data-driven method to assess virtual human animation quality. This approach achieves 90% correlation, outperforming deep learning baselines for animations not using parametric models.
Area of Science:
- Computer Graphics
- Human-Computer Interaction
- Virtual Reality
Background:
- Virtual human animations are crucial for virtual and augmented reality applications.
- Assessing the quality of automatically generated virtual humans is challenging.
- Existing quality assessment methods often rely on parametric body models, limiting their applicability.
Purpose of the Study:
- To introduce the first quality assessment measure for virtual human animations not generated with parametric body models.
- To develop a novel data-driven framework for evaluating animation realism.
Main Methods:
- Generated a dataset of virtual human animations.
- Collected subjective realism evaluation scores through a user study.
- Trained a linear regressor to predict perceptual evaluation scores using the dataset.
Main Results:
- Achieved a 90% correlation between predicted and subjective realism scores.
- The developed linear regressor significantly outperformed a strong deep learning baseline.
- Demonstrated the effectiveness of the data-driven framework for quality assessment.
Conclusions:
- The proposed data-driven framework provides an effective method for assessing virtual human animation quality.
- This approach is particularly valuable for animations not derived from parametric models.
- The study highlights the potential of machine learning for objective evaluation of complex visual content.
