基于多模态大模型的图片质量评估系统开发
Development of an Image Quality Assessment System Based on Multimodal Large Language Models
杜长星 1张冬梅1
作者信息
- 1. 北京邮电大学计算机学院，北京 100876
- 折叠
摘要
针对现有图像质量评估工具反馈单一、细粒度分析能力不足、缺乏成熟完整系统支持及模型适配性有限等问题，设计并实现一种融合自研微调模型与第三方开源模型的多模态细粒度图片质量评估系统。系统采用 C/S 与 B/S 混合架构及前后端分离设计，整合用户交互、业务处理、数据存储及多源 AI 模型推理等核心模块，支持多维度评分、可解释性评估报告生成、批量图像筛选、多模型适配切换等功能。通过显著物体识别、双图对比机制、动态损失函数优化及多模型协同推理策略，大幅提升细粒度曝光差异判别精度，同时保障系统的可靠性、易用性与扩展性。实际测试表明，系统在 KADID-10K、PIPAL 等基准数据集上表现优异，能有效满足专业摄影评估、摄影学习、商业图像筛选及日常图片管理等多场景需求，为图像质量评估提供高效、灵活、实用的解决方案。
Abstract
To address issues such as single-dimensional feedback, insufficient fine-grained analysis capabilities, lack of mature and complete system support, and limited model adaptability in existing image quality assessment tools, this study designs and implements a multimodal fine-grained image quality assessment system that integrates self-developed fine-tuned models and third-party open-source models.
The system adopts a C/S (Client/Server) and B/S (Browser/Server) hybrid architecture as well as a front-end and back-end separation design, integrating core modules including user interaction, business processing, data storage, and multi-source AI model inference. It supports functions such as multi-dimensional scoring, interpretable assessment report generation, batch image screening, and multi-model adaptation and switching.
Through strategies like salient object recognition, dual-image comparison mechanism, dynamic loss function optimization, and multi-model collaborative reasoning, the system significantly improves the accuracy of fine-grained exposure difference discrimination, while ensuring its reliability, usability, and scalability. Practical tests demonstrate that the system performs excellently on benchmark datasets such as KADID-10K and PIPAL, effectively meeting the needs of multiple scenarios including professional photography assessment, photography learning, commercial image screening, and daily image management. It thus provides an efficient, flexible, and practical solution for image quality assessment.关键词
图像质量评估/多模态大模型/细粒度分析/前后端分离/多模型融合/跨平台应用Key words
Image Quality Assessment/Multimodal Large Language Models/Fine-grained Analysis/Front-end and Back-end Separation/Multi-model Fusion/Cross-platform Application引用本文复制引用
杜长星,张冬梅.基于多模态大模型的图片质量评估系统开发[EB/OL].(2026-02-09)[2026-02-11].http://www.paper.edu.cn/releasepaper/content/202602-49.学科分类
计算技术、计算机技术
评论