DongSky/AEGIS

AEGIS: Exploring the Limit of World Knowledge Capabilities for Unified Multimodal Models [📂 GitHub] [🆕 Blog] [📜 Paper] Summary The capability of Unified Multimodal Models (UMMs) to apply world knowledge across diverse tasks remains a critical, unresolved challenge. Existing benchmarks fall short, offering only siloed, single-task evaluations with limited diagnostic power. To bridge this gap, we propose AEGIS (i.e., Assessing Editing, Generation… See the full description on the dataset page: https://huggingface.co/datasets/DongSky/AEGIS.