Huiyu Duan

MIEScore aligns human preferences in multi-source image editing, outperforming existing benchmarks and setting a new standard for evaluation in this complex domain.

LLMs with similar semantic skills show wildly different economic performance in simulated markets, revealing that reasoning about competition and resource allocation remains a major challenge.

Forget small-scale image editing datasets – this work unleashes a million-scale human preference dataset that unlocks better reward models and boosts text-guided image editing performance.