CLEAR: Unlocking Generative Potential for Degraded Image Understanding in Unified Multimodal Models

Xiangzhao Hao, Zefeng Zhang, Zhenyu Zhang +6

Image degradation from blur, noise, compression, and poor illumination severely undermines multimodal understanding in real-world settings. Unified multimodal models that combine u…