alpha-matte MFIF dataset

A large-scale training dataset suffering from the defocus spread effect (DSE) is synthesized by applying an $\alpha$-matte boundary defocus model to the VOC 2012 dataset.

Motivation: Due to the lack of large-scale datasets of multi-focus images, several data generation methods based on public natural image datasets have been adopted in many deep learning (DL)-based multi-focus image fusion algorithms. However, the DSE is neglected in all the abovementioned datasets. This unrealistic training data may limit the performance of these algorithms.

Application: For training DL-based multi-focus image fusion algorithms.