LMHaze
收藏资源简介:
LMHaze数据集由北京理工大学创建,是一个大规模、高质量的真实世界去雾数据集,包含5040对高分辨率的模糊和清晰图像,涵盖多种室内外环境和不同雾浓度。数据集大小超过现有最大真实世界去雾数据集的25倍,图像分辨率高达5472×3648。数据集创建过程中,通过专业的雾生成器和静态场景采集,确保了图像对齐和雾浓度的多样性。此外,数据集还提供了多类型的手工标注,包括目标检测、语义分割和图像描述,旨在评估去雾方法在下游任务中的性能,解决现有数据集在雾浓度分布和场景多样性方面的不足。
The LMHaze dataset, created by Beijing Institute of Technology, is a large-scale, high-quality real-world dehazing dataset. It contains 5040 pairs of high-resolution hazy and haze-free images covering diverse indoor and outdoor environments and varying fog concentrations. The dataset is over 25 times larger than the current largest real-world dehazing dataset, with an image resolution up to 5472×3648. During its creation, professional fog generators and static scene collection were employed to ensure image alignment and the diversity of fog concentrations. Furthermore, the dataset provides multiple types of manual annotations including object detection, semantic segmentation, and image captioning, aiming to evaluate the performance of dehazing methods in downstream tasks and address the shortcomings of existing datasets in terms of fog concentration distribution and scene diversity.

- 1LMHaze: Intensity-aware Image Dehazing with a Large-scale Multi-intensity Real Haze Dataset北京理工大学 · 2024年



