Rethinking Image Quality Assessment through the Lens of Task Utility in Embodied Settings

Abstract

Image quality assessment (IQA) underpins embodied imaging pipelines by judging whether visual quality satisfies downstream tasks, yet most methods learn task-agnostic scores aligned with generic human ratings on static benchmarks. This objective mismatches the embodied and interactive settings, where image adequacy depends on task goals, context, and action requirements that shape an agent’s decisions. We argue that IQA should shift from score regression to goal-conditioned judgment defined by the utility of embodied tasks. Such utility-aware assessment demands models with strong reasoning, grounding, and tool-use capabilities, as enabled by multimodal large language models (MLLMs) agent. We advocate rethinking IQA from the perspective of embodied task utility and outline benchmarks, evaluation protocols, and research directions for developing MLLM-based embodied IQA agents.