MattingMFIF
收藏资源简介:
MattingMFIF是一个新的4K合成数据集,用于多焦点图像融合(MFIF)的训练和验证。该数据集通过使用Blender软件从真实照片中模拟真实景深(DOF)效果创建而成,提供了10,000个样本,用于训练MFIF模型。数据集由Distinctions-646和PhotoMatte85数据集中的裁剪主体以及BK-20K数据集中的背景图像组成。通过使用Blender脚本准确放置相机和六个不同的平面,以及生成每个主体的渲染图像,模拟了各种MFIF示例。该数据集旨在解决MFIF训练数据稀缺的问题,为训练MFIF模型提供了一个真实的大规模基础。
MattingMFIF is a novel 4K synthetic dataset for training and validation of multi-focus image fusion (MFIF). It is constructed by simulating realistic depth-of-field (DOF) effects from real photographs using Blender software, providing 10,000 samples for training MFIF models. The dataset comprises cropped foreground subjects sourced from the Distinctions-646 and PhotoMatte85 datasets, alongside background images from the BK-20K dataset. Various MFIF examples are simulated by precisely placing the camera and six distinct planes via Blender scripts, and generating rendered images for each subject. This dataset aims to address the scarcity of training data for MFIF, providing a realistic and large-scale foundation for training MFIF models.



