Reasoning as Representation: Rethinking Visual Reinforcement Learning in Image Quality Assessment
This work addresses the deployment limitations of IQA models for scenarios requiring low latency and energy efficiency, offering a more practical solution.
The paper tackles the high computational cost of reasoning-based image quality assessment (IQA) models by identifying that their generalization stems from converting visual to text representations via reinforcement learning, and proposes RALI, which uses contrastive learning to achieve similar performance with less than 5% of parameters and inference time.
Reasoning-based image quality assessment (IQA) models trained through reinforcement learning (RL) exhibit exceptional generalization, yet the underlying mechanisms and critical factors driving this capability remain underexplored in current research. Moreover, despite their superior performance, these models incur inference energy usage and latency orders of magnitude higher than their earlier counterparts, restricting their deployment in specific scenarios. Through extensive experiments, this paper verifies and elaborates that through RL training, MLLMs leverage their reasoning capability to convert redundant visual representations into compact, cross-domain aligned text representations. This conversion is precisely the source of the generalization exhibited by these reasoning-based IQA models. Building on this fundamental insight, we propose a novel algorithm, RALI, which employs contrastive learning to directly align images with these generalizable text representations learned by RL. This approach eliminates the reliance on reasoning processes and even obviates the need to load an LLM. For the quality scoring task, this framework achieves generalization performance comparable to reasoning-based models while requiring less than 5% of their model parameters and inference time.