LHP-Rain
收藏资源简介:
LHP-Rain是由华中科技大学人工智能与自动化学院多谱信息处理技术国家级重点实验室创建的大规模高质量真实雨天图像数据集,包含3000个视频序列,总计超过100万对高分辨率(1920*1080)的雨天与晴天图像。该数据集旨在解决真实图像去雨(RID)中的关键问题,特别是针对多种雨类型,如雨丝、雾化效果、遮挡和地面溅水。数据集的创建过程中,通过智能手机捕捉高分辨率的真实雨视频,并采用新颖的鲁棒低秩张量恢复模型生成高质量的地面实况(GT)。LHP-Rain不仅用于雨天图像恢复,还支持在恶劣天气条件下的目标检测和分割任务,适用于自动驾驶和视频监控场景,旨在提高图像质量并解决真实世界中的雨天图像问题。
LHP-Rain is a large-scale, high-quality real rainy image dataset developed by the National Key Laboratory of Multispectral Information Processing Technology, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology. It consists of 3000 video sequences, with a total of over 1 million pairs of high-resolution (1920*1080) rainy and clear-sky images. This dataset aims to address the core challenges in real-world image deraining (RID), especially covering diverse rain types including rain streaks, fog-like artifacts, occlusion, and ground splashing. During the dataset construction, high-resolution real rainy videos were captured using smartphones, and a novel robust low-rank tensor recovery model was adopted to generate high-quality ground truth (GT). Beyond rainy image restoration, LHP-Rain also supports object detection and segmentation tasks under adverse weather conditions, which is applicable to autonomous driving and video surveillance scenarios. Its goal is to improve image quality and solve real-world rainy image-related problems.




