Reproduce “Quality assessment of 3D human animation: Subjective and objective evaluation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rim Rekik, Stefanie Wuhrer, Ludovic Hoyet, Katja Zibrek, Anne-Hélène Olivier
This research introduces 4DHumanPercept, the first dataset of perceptually rated, systematically distorted 3D human animations, and proposes 4DHumanQA, a data-driven objective metric. 4DHumanQA predicts human perception of animation quality for non-parametric geometrically dense virtual humans, achieving a Spearman Rank Order Correlation Coefficient of 0.961 and outperforming deep learning baselines.
Key insights
- Perceived animation quality is highly influenced by distortion strength, with complex interactions between VH characteristics (identity, clothing, motion type) and observer factors affecting perceived degradation.
- No single objective feature alone adequately captures overall human perception of animation realism, indicating the necessity for a multi-feature, combined approach.
- A perceptually optimized linear model, leveraging carefully selected objective features, can outperform a general image-based deep learning metric (LPIPS) for assessing 3D human animation quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.