Hello darkperson8,
I can suggest to generate the masks from the model (in the duplicated chunk) generated from the sparse cloud, if it has the valid shape even a rough one, then re-align the image set and rebuild the dense cloud.
For alignment I would suggest High accuracy, Generic preselection, 40 000 key points, 10 000 tie points, adaptive fitting - disabled. For dense cloud - Mild filtering. If you have Store depth option in the Advanced preferences tab enabled, you can re-use the depth maps and generate the mesh model based on the depth maps (I recommend to use version 1.6.0 pre-release for it).