CLEVR is a synthetic diagnostics dataset for compositional language and elementary visual reasoning — consisting of simple geometric shapes (cylinders, cubes, spheres) rendered with physics engines to test pure logical reasoning without the complexity of real-world visual noise.

What Is CLEVR?

• Definition: Compositional Language and Elementary Visual Reasoning.
• Visuals: Clean, rendered scenes of geometric objects.
• Questions: Highly complex, nested logic (e.g., "What is the shape of the object that is nearest to the tiny yellow cylinder?").
• Goal: Isolate reasoning ability from visual recognition difficulty.

Why CLEVR Matters

• Solved by Modules: Spurred the development of Neuro-Symbolic AI and Modular Networks.
• Proof of Logic: If a model fails CLEVR, it cannot reason, even if it recognizes faces perfectly.
• Benchmarking: Standard test for "System 2" visual thinking.

CLEVR is the playground for visual logic — a simplified universe used to teach AI how to think step-by-step.