Benchmark RadarCRONOS
CRONOS is an intervention-based benchmark for evaluating counterfactual physical consistency in video prediction models. It provides a photorealistic Unreal Engine environment with controlled videos of physical events (collision, occlusion, fall) while intervening on viewpoint, scene, object category, and object appearance. The evaluation protocol defines six metrics computed via video segmentation, tracking, 3D reconstruction, and VLM-based task performance.
- Released
- 2026-05-22
- Readiness
- Runnable
- Primary field
- Cybersecurity
Why it matters
Current video models often rely on superficial correlations rather than causal structure. CRONOS enables systematic diagnosis of how prediction quality degrades under controlled interventions, providing a concrete target for developing models robust to variations in viewpoint and context.
Motivation
Video prediction is increasingly viewed as a path toward generalizable world models, yet it remains unclear whether these systems learn underlying causal structure or merely exploit superficial visual correlations for future prediction.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.