Pointcloud: Tackling Noise

[Alexey Pasumansky]

In the version 1.5.0 we implemented new algorithm of model reconstruction from depth maps which is faster and leads to better quality mesh. It is able to build model from depth maps (from 1421 photos) in 5 hours 40 minutes (in addition to this time the depth maps have been generated within 5 hours 30 minutes).
Computer specification used for this test: AMD Ryzen Threadripper 2950X + 3 x NVIDIA GeForce GTX 1080 Ti + 64 Gb RAM.

You can check quality in attached screenshots in solid and shaded view modes (shaded mode displays the interpolated colors of the mesh vertices, so it is not textured here).

[Mak11]

In the version 1.5.0 we implemented new algorithm of model reconstruction from depth maps which is faster and leads to better quality mesh. It is able to build model from depth maps (from 1421 photos) in 5 hours 40 minutes (in addition to this time the depth maps have been generated within 5 hours 30 minutes).
Computer specification used for this test: AMD Ryzen Threadripper 2950X + 3 x NVIDIA GeForce GTX 1080 Ti + 64 Gb RAM.

You can check quality in attached screenshots in solid and shaded view modes (shaded mode displays the interpolated colors of the mesh vertices, so it is not textured here).

That's a monster configuration! But I can confirm that the new reconstruction method is a lot faster that the default method (dense cloud + mesh generation) and that it generates more precise models (and requires less RAM!).

The team did a great job (alignement is now also a lot faster when there's lots of overlap).

Mak