I've been thinking about photogrammetry and 3dgs lately. One consideration is the depth of field of images. Afaik it is somewhat common to focus stack images, or just use small apertures, before feeding them to photogrammetry/3dgs pipeline. But that doesn't really make sense because focus stacking is lossy operation and requires some way to estimate which parts are in focus. So it would make more sense to have the photogrammetry/3dgs work directly on unstacked shallow dof images. Ideally the algorithm should use the focus information to enhance the reconstruction process somehow.
