MVSR4×
收藏资源简介:
MVSR4×是由哈尔滨工业大学创建的实际视频超分辨率数据集,专门设计用于解决现有VSR方法在处理复杂退化真实世界数据时的挑战。该数据集通过智能手机的不同焦距镜头同步捕捉低分辨率和高分辨率视频,确保了数据多样性。数据集包含300个高质量的LR-HR序列,每个序列有100帧,分辨率为1080P。MVSR4×不仅在数据采集过程中应用了色彩校正和位置对齐算法,还在训练阶段进行了优化,以减少对VSR模型学习的不利影响。该数据集适用于开发和测试针对真实场景的视频超分辨率技术,旨在提高VSR方法在实际应用中的性能和效果。
MVSR4× is a real-world video super-resolution (VSR) dataset created by Harbin Institute of Technology, specifically developed to address the challenges encountered by existing VSR methods when handling real-world data with complex degradations. This dataset synchronously acquires low-resolution (LR) and high-resolution (HR) videos via smartphone lenses with varying focal lengths, ensuring comprehensive data diversity. The dataset consists of 300 high-quality LR-HR video sequences, each comprising 100 frames at 1080P resolution. MVSR4× not only applies color correction and spatial alignment algorithms during the data collection process, but also optimizes the preprocessing pipeline during the training stage to mitigate adverse impacts on the learning of VSR models. This dataset is suitable for developing and testing real-world video super-resolution technologies, with the goal of improving the performance and practical effectiveness of VSR methods in real-world applications.

- 1Benchmark Dataset and Effective Inter-Frame Alignment for Real-World Video Super-Resolution哈尔滨工业大学, 中国 · 2022年



