Overview
Morpheus is a benchmark with real-world videos of physical phenomena that obey conservation laws (mass, energy, momentum). It uses physics-informed metrics, neural networks, and VLMs to evaluate whether generative models respect these laws.Why it matters
Unlike synthetic benchmarks, Morpheus grounds evaluation in real physics. It highlights where models violate physical consistency, essential for robotics, simulation, and engineering uses.Key trade-offs / limitations
- Benchmark scope limited to curated physical setups.
- Real data is harder to scale than synthetic benchmarks.