Hi,
I concur with this, I am seeing poorer 3D modelling directly from depth maps. I am processing an airborne imagery of an industrial facility with overhead conveyor structures. Processing a point cloud in ultra-fine/mild filtering reconstructs these overheard objects, but with same settings for depthmap based 3D model, very poorly reconstructs these objects, with large gaps in these features. I am happy to provide the source data-set to Agisoft to assist development.
Cheers,
Stu.