Messages

1

General / Re: Tiled Model Banding Problem

« on: May 02, 2022, 11:08:25 AM »

Hi,

I've now tested the same project within Agisoft Cloud. Unfortunately the problem is still present. I think this rules out any hardware/driver issues. I have shared the produced tiled model with support@agisoft.com.

I'm starting to think this is a software bug?

I'm happy to further test any tweaks/settings though, perhaps it's a CRS issue?

Many thanks,
Paul

2

General / Re: Tiled Model Banding Problem

« on: April 26, 2022, 05:14:08 PM »

Hi Alexey,

System specification:

Windows 10 Pro 19043
Intel i7-8700
32GB DDR4
RTX-2070 8GB  Driver version 512.15
Metashape Pro Version 1.8.3 64bit

I don't have a copy of the original logs but I have reprocessed a small region at a point where the problems were present and have attached the log below. (This smaller reprocessed region also showed the exact same issues as in the full tiled model).

I have found that if I increase the tile size to 4096 then the whole model fits on to one of the large squares/tiles/cubes, and it all looks fine. It's obviously much slower to load like this but it looks good without any of the issues.  Please see attached images, tile size 4096 looks fine, but tile size 256 looks bad.  Probably the larger tile size has just scaled out the problem to a larger geographic region? Either way it shows the dataset is not the problem.

The 4096 tile size model has consistent lighting across the whole model, and no missing small squares.

The attached image "squares.jpg" shows a more zoomed out look at the 256 tiled model.

I upgraded my GPU driver to the latest version (512.15) as I had a much older version installed, but this made no difference.

I have also tested on Metashape version 1.8.1 and the problem is still present.

Many thanks in advance for your help!
Paul

3

General / Re: Tiled Model Banding Problem

« on: April 25, 2022, 11:23:09 PM »

I've now tried producing a tiled model using all three sources 1) Depth Maps, 2)Dense Cloud, 3) Mesh, and I'm still get big square blocks of different shading, and also missing blocky areas in black (which I know I have good data for as it worked fine in the smaller samples).

It seems that the blocks are in North to South and East to West direction.  The small blocks look to be the tile size of 256, and the larger blocks seem to be composed of 16x16 smaller blocks.  Each block is apparent as the shading is a bit different to adjoining blocks, and in some places there are many small blocks missing from the render.

Not sure where to go from here, any advice greatly received :-)

Many thanks,
Paul

4

General / Tiled Model Banding Problem

« on: April 24, 2022, 12:16:47 PM »

Hi,

I've been experimenting with a large dataset, trying to learn Metashape.  I'm trying to produce a full tiled model of the dataset and have run into a problem with square cross banding and black holes (please see attached image).  The banding seems to be across the whole area and doesn't match with any flight lines, it seems this is being produced by the processing.

My workflow was generally to:

Alignment  -> produce dense cloud (in high, without color) -> produce DEM and orthomosaic -> produce mesh (in high, without color) - > produce tiled model based on mesh source (in high).

I did have quite a few photos unticked in the reference window as the GPS quality was affecting the accuracy of the alignment, but the mesh model seems very good.

 I have already produced very good tiled models from smaller regions of the dataset that worked just fine and without any banding. For these sets I just produced the tiled model straight after producing the dense cloud and selected the dense cloud as source.

I will try to run the tiled model again tonight based on dense cloud rather than the mesh, but I wonder if anyone has run into this problem before and has any tips?

Many thanks,
Paul

5

General / Re: Roof problem

« on: April 02, 2022, 06:12:12 PM »

Thanks for your reply,

I took off from a flat roof which was about the same height as the average height of the roofs. I set the altitude to 15M with a front and side overlap of 75%.  From looking at the photos manually I do seem to have achieved this over most of the scene, however at the peak roof height (which I think was about 5M below the drone) it falls to 40% sidelap and 50% frontlap. This is roughly what I estimated as I assumed this would still work out OK for producing a 2.5D orthomosaic.

I have re-run the dense cloud with "Calculate point confidence" turned on and have filtered the dense cloud by confidence. As I increase the minimum confidence, lower parts of the roof do start disappearing, so I presume this does mean it's an overlap problem?

If so, the next step would be to re-fly over the roof a bit higher and ensuring more overlap? And then I could re-insert those photos into the project and mask out any ground areas where things may have changed?

Many thanks again for your help!

Paul

6

General / Roof problem

« on: April 01, 2022, 07:22:31 PM »

Hello,

I've recently been trialling Metashape and have used my DJI Mini 2 to capture a typical scenario.  However I'm having some trouble with one of the roofs in the scene. This roof is higher than all the others which I think might be part of the problem. 

I have:

• Tried using DNG's rather than the JPG's
• Tried various levels of cloud density
• Tried the 3 different depth filtering values Mild/Moderate/Aggressive
• Tried manually aligning the photos from along the ridge of the roof with hundreds of tie points
• Tried a chessboard camera calibration

Unfortunately nothing seems to fix it.  The photos taken above the ridge of the roof seem sharp and clear, and Metashape has no problem aligning them. Please see attached images.

Any help or advice would be greatly appreciated!

Many thanks,
Paul