Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Topics - Alexandros21

Pages: [1]
General / Change in hardware leads to sparser depth maps
« on: March 01, 2021, 12:43:39 PM »
Hi all,

I've started using a Cloud service for my work and I also installed Metashape there. I ran a dataset both on the cloud and my local machine, using the exact same Metashape version (1.7.1. build 11797) and the exact same tweaks (BuildDenseCloud/max_neighbors = -1).

However, the generated depth maps in the Cloud are sparser than the depth maps built in the local machine and consequently, they lead to a much sparser pointcloud. Could you please help me understand what might be the issue? Could it be eg. because of the different GPU? Here is the hardware that I've used:

Local machine:
Processor Intel Core i7-7700k CPU @ 4.2 GHz
GeForce GTX 1080 (20 compute units @ 1733 MHz, 8192 MB)

AMD EPYC 7V12 64-core @ 2.44 GHz
RAM 112 GB
Radeon Instinct MI25 MxGPU (gfx900) (64 compute units @ 1000MHz, 16064 MB)

Thank you in advance,

Hi everyone,

I've created a high quality mesh model from a building with big glass structures, using Metashape 1.6. I know that the recommended way to run mesh is with mild filtered depth maps, however I get massive holes wherever there is glass. Oddly, if I use aggresive depth maps filtering the glass gets reconstructed much better, but the rest model is not as smooth.

Between 1.6.1 and 1.6.2 I also noticed a worsening of my mesh model - the surface is less smooth and the lines are less straight.

Is there any way of improving mesh model from depth maps with mild filtering? I even tried in 1.6.3 but I still get holes in the glass.

If not, is there at least a way to go back to 1.6.1's mesh model generation?

I can get slightly better results by using the visibility consistent method tweak, but I'd prefer to avoid this method, since it's really RAM-expensive.

Any help is much appreciated!

Thank you in advance,

Pages: [1]