Q-Router: Agentic Video Quality Assessment with Expert Model Routing and Artifact Localization

2025-10-09

AI Summary

Current video quality assessment (VQA) models suffer from poor generalization under direct score supervision, limited interpretability, and difficulty adapting to emerging content types—including user-generated content (UGC), short videos, and AI-generated content (AIGC). To address these limitations, we propose Q-Router, a general-purpose VQA framework based on multi-level agent-style routing. It dynamically selects and weights multiple specialized expert models using vision-language models, enabling spatiotemporal artifact localization and real-time inference. Crucially, Q-Router formalizes routing as an interpretable decision process, facilitating content-adaptive quality prediction. Experiments demonstrate that Q-Router achieves state-of-the-art performance across multiple VQA benchmarks, significantly improving cross-dataset generalization. It also excels on the Q-Bench-Video question-answering task and validates its efficacy as a reward function for post-training video generation models.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Planning, Routing, and Scheduling: Planning with Language Models
• Knowledge Representation and Reasoning: Qualitative Reasoning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Web Mining and Content Analysis: Web data quality in the era of algorithmically-generated content

Abstract

Video quality assessment (VQA) is a fundamental computer vision task that aims to predict the perceptual quality of a given video in alignment with human judgments. Existing performant VQA models trained with direct score supervision suffer from (1) poor generalization across diverse content and tasks, ranging from user-generated content (UGC), short-form videos, to AI-generated content (AIGC), (2) limited interpretability, and (3) lack of extensibility to novel use cases or content types. We propose Q-Router, an agentic framework for universal VQA with a multi-tier model routing system. Q-Router integrates a diverse set of expert models and employs vision--language models (VLMs) as real-time routers that dynamically reason and then ensemble the most appropriate experts conditioned on the input video semantics. We build a multi-tiered routing system based on the computing budget, with the heaviest tier involving a specific spatiotemporal artifacts localization for interpretability. This agentic design enables Q-Router to combine the complementary strengths of specialized experts, achieving both flexibility and robustness in delivering consistent performance across heterogeneous video sources and tasks. Extensive experiments demonstrate that Q-Router matches or surpasses state-of-the-art VQA models on a variety of benchmarks, while substantially improving generalization and interpretability. Moreover, Q-Router excels on the quality-based question answering benchmark, Q-Bench-Video, highlighting its promise as a foundation for next-generation VQA systems. Finally, we show that Q-Router capably localizes spatiotemporal artifacts, showing potential as a reward function for post-training video generation models.

Problem

• Improving generalization across diverse video content types
• Enhancing interpretability through artifact localization mechanisms
• Increasing extensibility to novel video quality assessment tasks

Innovation

• Agentic framework with multi-tier expert model routing
• Vision-language models dynamically route appropriate expert models
• Spatiotemporal artifact localization for improved interpretability