UniPercept: Towards Unified Perceptual-Level Image Understanding across Aesthetics, Quality, Structure, and Texture

Multimodal large language models (MLLMs) have achieved remarkable progress in visual understanding tasks such as visual grounding, segmentation, and captioning. However, their ability to perceive perceptual-level image features remains limited. In this work, we present UniPercept-Bench, a unified framework for perceptual-level image understanding across three key domains: Aesthetics, Quality, Structure and Texture. We establish a hierarchical definition system and construct large-scale datasets to evaluate perceptual-level image understanding. Based on this foundation, we develop a strong baseline UniPercept trained via Domain-Adaptive Pre-Training and Task-Aligned RL, enabling robust generalization across both Visual Rating (VR) and Visual Question Answering (VQA) tasks. UniPercept outperforms existing MLLMs on perceptuallevel image understanding and can serve as a plug-andplay reward model for text-to-image generation. This work defines Perceptual-Level Image Understanding in the era of MLLMs and, through the introduction of a comprehensive benchmark together with a strong baseline, provides a solid foundation for advancing perceptual-level multimodal image understanding.

Q: What are the people looking at near the shore? A: The sunset over the ocean.

A: The image shows a grand cathedral with intricate spires and stone carvings under a clear blue sky, surrounded by people in the open square.

A: The central placement of the cat is a key factor in achieving visual balance in this image. The cat acts as the focal point, drawing immediate attention and anchoring ...

A: The image depicts a surface with a crumpled texture, characterized by irregular folds, creases, and a matte finish. The characteristics ...