Out of Sight, Out of Mind? Evaluating State Evolution in Video World Models
Paper • 2603.13215 • Published
Search is not available for this dataset
image imagewidth (px) 1.28k 7.88k |
|---|
End of preview. Expand in Data Studio
STEVO-Bench
STEVO-Bench evalutes whether video world models can “simulate” a world beyond the pixel frame. We introduce observation control such as occlusion, camera lookaway or illumination dimming to prove whether video world models can evolve state successfully Visit the STEVO-Bench official site to see leaderboard, example tasks and example videos.
Read our paper and blog post to see our findings and discussions about SOTA video world models!
📄 Paper | 📝 Blog | 🌐 Website | 💻 Code
Usage
Check out the STEVO-Bench github repo for full instructions on using the benchmark.
Task Categories
STEVO-Bench includes 225 tasks from 6 different categories, each of which captures a fundamentally distinct mode by which the world changes. The categories span scalar processes that accumulate gradually over time, simple few-object kinematics, irreversible structural or material transformations, and intent-driven behavior of agents etc. Together, they cover the physical, chemical, and social dimensions of world dynamics, providing a comprehensive probe of a model's internal simulation capacity.
- Downloads last month
- 40
Size of downloaded dataset files:
394 MB
Size of the auto-converted Parquet files:
394 MB
Number of rows:
244