doanh25032004apache-2.0

Video World Model Acceleration - Qualitative Quality Ladder

A qualitative benchmark dataset showing the degradation of a video world model (Cosmos-Predict2 2B DiT) as WorldCache diffusion caching is progressively applied, from quality-preserving to complete collapse. Includes single-view and multi-view robot manipulation videos with detailed fidelity metrics.

Downloads35
Episodes8

Why This Matters for Physical AI

This dataset demonstrates the trade-offs between inference speed and visual fidelity in video world models for robotics, crucial for understanding practical deployment constraints of generative models in embodied AI systems.

Technical Profile

Modalities
rgb
Robot Embodiments
humanoid
Environment
lab
Task Types
manipulation
Episodes
8
Data Format
mp4
Annotation Types
quality_metrics
License
apache-2.0
Part of the Video World Model Acceleration family

Access

Need custom rgb data?

Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets