Gaps after build dense cloud

[Jakker]

Hi there!

I'm currently trying to build a dense cloud of a river bank for 5 different time frames using the 1.7.1 PRO version.
I have masked the images before alignment and aligned using the high setting with a 40k/10k key/tie point limit.
The generated sparse cloud does look mostly very good for all timeframes, using around 40-60 images per timeframe from 6-7 different camera positions.
I then deleted the noise pixels in the background and built a dense cloud in high setting.
What I found was that my created dense clouds, trying all different depth filtering settings, all have either small or very large gaps.
Sometimes the sparse cloud is even higher in detail than the dense cloud.
Do you have any idea how I can build a dense cloud with more accuracy?
I have been trying to play around with the gradual selection panel deleting some points, but the dense cloud does not increase in accuracy.
Below I have attached some images to display my issue, they were done using the medium accuracy/quality but its the same there.

One more thing I'd like to add. I have used the older version photoscan software some years ago and remember that the dense cloud generation was a lot more accurate.


Thanks in advance!
Regards
Jakob

[Alexey Pasumansky]

Hello Jakob,

Can you send the source images and the project with the alignment results to support@agisoft.com?

[Jakker]

Hi Alexey!

Will do. Thanks for your quick reply.

Regards
Jakob

[Alexey Pasumansky]

Hello Jacob,

Thank you for sharing the sample dataset.

I have split all the camera calibration groups in the Tools Menu -> Camera Calibration dialog, then optimized the camera alignment having all the parameters (except for B1 and B2) enabled and then re-built the dense point cloud using High quality and Mild filtering options.

This results in somewhat more dense point cloud compared to your results. Please check, if you can get the desired output following the same steps.


If the images have been captured using the nodal head, then you could group them to the camera station folders in the Workspace pane.

[Alexey Pasumansky]

Seems that the problem is partially related to the 1.7.1 depth maps generation issue related to the masks applied to some images.

In the 1.7.2 pre-release the problem is solved and the dense cloud contains reasonable number of points.