Scholarlink

<p>Recently, User-Generated Content videos have gained popularity in our daily lives. However, these videos often suffer from poor exposure due to the limitations of photographic equipment and techniques. Therefore, video exposure correction algorithms have been proposed, including low-light video enhancement and over-exposed video recovery. Equally important to the video exposure correction is the Video quality assessment (VQA), which aims to objectively evaluate the perceptual quality of enhanced videos. In this work, we introduce the <b>VEC-QA</b> dataset, a large-scale VQA benchmark comprising both low-light and over-exposed videos along with their corrected counterparts. Furthermore, we propose <b>Light-VQA+</b>, a video quality assessment model specialized in assessing the perceptual quality of video exposure correction results. Light-VQA+ is built upon a vision-language foundation by integrating the CLIP model and vision-language guidance during feature extraction. It also incorporates a novel module inspired by the Human Visual System to achieve more accurate and perceptually aligned quality predictions. Extensive experimental results show that our model achieves the best performance against the current state-of-the-art VQA models on the VEC-QA dataset and other public datasets. Our code and dataset can be found at <a href="https://github.com/SaMMyCHoo/Light-VQA-plus">https://github.com/SaMMyCHoo/Light-VQA-plus</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Light-VQA+: A Video Quality Assessment Model for Exposure Correction with Vision-Language Guidance

  • 期刊论文

  • 发表日期：

    2026年3月6日

查看全文
  • Xunchu Zhou,
  • Xiaohong Liu,
  • Yunlong Dong,
  • Tengchuan Kou,
  • Yixuan Gao,
  • Zicheng Zhang,
  • Chunyi Li,
  • Haoning Wu,
  • Guangtao Zhai

摘要

Recently, User-Generated Content videos have gained popularity in our daily lives. However, these videos often suffer from poor exposure due to the limitations of photographic equipment and techniques. Therefore, video exposure correction algorithms have been proposed, including low-light video enhancement and over-exposed video recovery. Equally important to the video exposure correction is the Video quality assessment (VQA), which aims to objectively evaluate the perceptual quality of enhanced videos. In this work, we introduce the VEC-QA dataset, a large-scale VQA benchmark comprising both low-light and over-exposed videos along with their corrected counterparts. Furthermore, we propose Light-VQA+, a video quality assessment model specialized in assessing the perceptual quality of video exposure correction results. Light-VQA+ is built upon a vision-language foundation by integrating the CLIP model and vision-language guidance during feature extraction. It also incorporates a novel module inspired by the Human Visual System to achieve more accurate and perceptually aligned quality predictions. Extensive experimental results show that our model achieves the best performance against the current state-of-the-art VQA models on the VEC-QA dataset and other public datasets. Our code and dataset can be found at https://github.com/SaMMyCHoo/Light-VQA-plus.