PDI-Bench: Quantitative Video World Model Evaluation for Geometric Consistency
Paper Overview
- Field: Computer Vision (CV)
- Authors: Jiaxin Wu, Yihao Pi, Yinling Zhang, Yuheng Li, Xueyan Zou
- Published: 2026-05-14
- arXiv: 2605.15185
- Scale-depth alignment
- 3D motion consistency
- 3D structural rigidity
- Across state-of-the-art video generators, PDI reveals common geometry-specific failure modes that are not captured by common perceptual metrics.
- PDI provides a diagnostic signal for progress toward physically grounded video generation and physical world models.
Abstract
Generative video models are increasingly studied as implicit world models, yet evaluating whether they produce physically plausible 3D structure and motion remains challenging. Most existing video evaluation pipelines rely heavily on human judgment or learned scorers, which can be subjective and weak at diagnosing geometric failures.
The authors introduce PDI-Bench (Perspective Displacement Index), a quantitative framework for auditing geometric consistency in generated videos. Given a generated clip, the framework:
1. Obtains object-centric observations through segmentation and point tracking (e.g., SAM 2, MegaSaM, and CoTracker3) 2. Lifts these observations into 3D world-space coordinates via monocular reconstruction 3. Computes a set of projected geometric residuals capturing three failure dimensions:
To support systematic evaluation, the authors construct the PDI dataset, covering diverse scenes designed to test these geometric constraints.
Findings
*Automatically collected on 2026-05-17*