Revisiting MLLM Based Image Quality Assessment: Errors and Remedy

The rapid advancement of multi-modal large language models (MLLMs) has significantly impacted image quality assessment (IQA), yet challenges persist due to the mismatch between discrete token outputs and the continuous quality scores needed for effective evaluation. Previous methods converting these outputs often resulted in errors, limiting the performance of MLLM-based IQA. To address this, the new framework Q-Scorer has been proposed, which integrates a lightweight regression module and IQA-specific score tokens into the MLLM pipeline. Extensive experiments have shown that Q-Scorer achieves state-of-the-art performance across multiple IQA benchmarks, demonstrating its ability to generalize well to mixed datasets. This development is crucial as it not only resolves existing issues but also enhances the overall effectiveness of MLLMs in various AI applications, paving the way for improved image quality assessments.