EBB!

该数据集名为EBB!，包含了在不同光照和天气条件下拍摄的散景（Bokeh）和无散景图像的对。该数据集共包含5094对图像，其中4694对用于训练，各有200对用于验证和测试。所有图像均使用窄（f/16）和宽（f/1.8）光圈设置拍摄，平均分辨率为1024×1536。在训练过程中，还应用了数据增强技术。这是一个大尺度数据集，含有5094对图像，其任务是仅从单一图像实现散景效果渲染。

The dataset, named EBB!, comprises pairs of bokeh and non-bokeh images captured under diverse lighting and weather conditions. In total, this dataset contains 5094 image pairs, with 4694 pairs allocated for training, and 200 pairs each reserved for validation and testing. All images were captured using both narrow (f/16) and wide (f/1.8) aperture settings, with an average resolution of 1024×1536. Data augmentation techniques were employed during the training process. This is a large-scale dataset featuring 5094 image pairs, whose core task is bokeh effect rendering from a single input image alone.

搜集汇总

背景与挑战

背景概述

EBB!数据集包含5000对宽/浅景深图像对，用于训练深度学习模型实现移动设备上的实时逼真散景渲染。数据集支持竞赛，要求模型在PSNR和延迟之间取得最佳平衡。