I had some time to test a few datasets, mainly captures by UAV's and Metashape really has a problem in dense clouds. Usually they are a lot poor than the ones produced in version 1.44. Even with the filtering disabled, lots of areas disappear from the cloud leaving huge holes. This problem is common mainly in areas with vegetation or more complex terrain. The last version usually solves these areas quite well. Metashape tends to eliminate them, apparently with excessive filtering. It's not a case of a single dataset or special cases. Its consistent whenever there's terrain and vegetation. The tied model improves things a bit because these areas get solved. However there are a few things that still need a dense cloud to be generated (for instance point classification). Meshes from dense clouds are also more detailed, at the present stage, than meshes produced using tiled model. Probably things will improve if some bugs in high quality tiled model generation are solved and a DEM can be generated from it.
Cheers
PS: Tried with filtering disabled and the cloud, as expected, does not have any holes. It seems that the thresholds for the filtering limits have been tighten, maybe too much. I would say that the depth Mild filtering in now equivalent to the Moderate option in PS 1.4.4 and "Disabled" from MS is almost equivalent and Mild in PS 1.44. It's more or less what James_AU is also saying in the next post. In UAV surveys, we might need the data that is "disappearing" now in Metashape if Mild Filter is used but using disabled is a little bit excessive also. Maybe we could get a few more options in the middle to get the best of both versions.