Agisoft Metashape 1.7.0 pre-release

[Alexey Pasumansky]

Hello Paul,

I've made a small comparison for the depth maps generation time on the single machine (i7-4960X, GeForce RTX 3080, Windows 7) for 650 DJI nadir photos in High quality with Mild filtering.

1.7.0.11736
Depth maps:   47 min
Memory peak: 7.4 GB

1.6.6.11715
Depth maps:    65 min
Memory peak:  2.7 GB

[Paulo]

Hi Alexey,

obviously with more recent GPUs processing times for recent version are better.  But I am stuck with an older GeForce GTX 670MX and for this GPU times for depth mas generation are much longer!

Have you been able to investigate why I get much more noise in dense cloud for latest version?

[Alexey Pasumansky]

Hello Paul,

The noise filtering will be improved with the next update.

As for the performance drop on the old cards, not sure, if we can improve it considerably.

[3D_Scan_Fan]

Also, any version after that is not compatible with Photoscan 1.3.5. because of the removal of Spir-1.2.

Funny enough, the last time I saw this similar problem was with newer version of the driver.

Hello 3D_Scan_Fan,

Metashape 1.7 uses hardware-accelerated textured from OpenCL API, which seems to be improperly supported in the drivers that you are using.

Is there any reason, why you need to use PhotoScan 1.3?

Every version from 1.6.5. and downwards work fine with the drivers I'm using.

As for Photoscan 1.3.5., I've probably explained to you in another post and in emails that there are aspects in that version (particularly the Alignment process) tends to be better.

I don't understand why now with 1.7 it gets broken.

[Paulo]

Hello Paul,

The noise filtering will be improved with the next update.

As for the performance drop on the old cards, not sure, if we can improve it considerably.

Thanks, Alexey for the update. I am now redoing the dense cloud using following tweak main/depth_pm_enable false and get results similar to 1.6 version see attached screen grab....

For now,I will be using this tweak on my laptop.

[3D_Scan_Fan]

I'm still getting the Error: ciErrNum: CL_BUILD_PROGRAM_FAILURE (-11) at line 1087 when trying to process Depth Maps

Might this be fixed by any chance?? 

[Paulo]

Hi all,

 I have a model constructed from 188 aerial drone images referenced to Coordinate system   WGS 84 + EGM96 height.

When I try to export to KMZ format I get Error: Vertical datum out of range

Why is this so?

NB. the export to KMZ works  in version 1.6.6

[Andreas1206]

I have a NVIDIA GeForce GTX 1660 Ti (6GB ram) and a NVIDIA RTX 2080 Super (8GB ram) in my system - both activated in Metashape. In version 1.65 they always shared the GPU workload but in version 1.7 I alawys get the following message in the log:

2021-01-11 11:39:52 Warning: Device GeForce GTX 1660 Ti ignored because it has 4572/6144 MB available, but 6134 MB required

The RTX 2080 Super is used. Does this mean, different GPUs with different amounts of RAM cannot be used together anymore in the new depth map generation process of version 1.7?

thanks

[eltama]

Hello Andreas
Try to connet RTX 2080 S to your monitor, so the GTX 1660 VRAM is completly free, close all other apps and lower monitor resolution.
Hope u fix it.
Regards
Elmar 

[Andreas1206]

Hello Elmar,

thank you for your reply - unfortunately this did not solve the problem. I also tried the "limiting neighbors" - tweak as Alexey recommended in an earlier post. At first none of the GPUs was used in the process. Limiting the neighbors to 40 with the tweak at least the RTX 2080 Super is used - but the 1660 Ti is constantly ignored. Is it maybe possible, that the algorithm is fully utilizing the capacity of the more powerful GPU, thus not being able to use the 1660 Ti anymore?

best regards,
Andreas

[Mak11]

Alexey,

1.7 really does haves serious issues with floating geometry inside & outside meshes and other anomalies on the edges.

From the Banana project:


One more try with another project:


Just came across those from an other user on Facebook ( the blue one is from metashpae 1.7):




Mak

[Alexey Pasumansky]

I have a model constructed from 188 aerial drone images referenced to Coordinate system   WGS 84 + EGM96 height.

When I try to export to KMZ format I get Error: Vertical datum out of range

Why is this so?

NB. the export to KMZ works  in version 1.6.6

Hello Paul,

We'll fix this problem in the next 1.7 update.

[Alexey Pasumansky]

I have a NVIDIA GeForce GTX 1660 Ti (6GB ram) and a NVIDIA RTX 2080 Super (8GB ram) in my system - both activated in Metashape. In version 1.65 they always shared the GPU workload but in version 1.7 I alawys get the following message in the log:

2021-01-11 11:39:52 Warning: Device GeForce GTX 1660 Ti ignored because it has 4572/6144 MB available, but 6134 MB required

The RTX 2080 Super is used. Does this mean, different GPUs with different amounts of RAM cannot be used together anymore in the new depth map generation process of version 1.7?

Hello Andreas,

Can you please share the complete log related to the depth maps calculation stage? Also please specify, which type of the images you are working with (resolution, file type and bit depth).

[Andreas1206]

Hello Alexey,

thank you for your reply. Attached you find a log file with the mentioned problem.

I usually work with .dng files, converted from Nikon D850 .nef Raw files - so a resolution of 8256x5504 pixels and a bit depth of 14bit.

thanks for your help,

Andreas

[ChristianeZhao]

Had you try to use a more precise panorama camera? we have tried the new 1.7.0 with photos captured by Civetta and the result is pretty good. the new upgrade of Agisoft is really a thing.

In case you are also interested in Civetta: https://weiss-ag.com/civetta360camera/

[For laser scans] Current approach is the following:
...
- run Align Photos operation as usual.

Does that mean that the poses of the laser scans will be adjusted in addition to the internal and external parameters calculated for the standard cameras?

As a user who commonly combines laser scan data with photogrammetry that is not at all what I would want to happen. The laser scan data will almost always be coming from other software where we have much more control over the alignment process. Therefore the laser scan data should be treated as fixed (no pose adjustments) and photos aligned to it.

I haven't had a chance to test the implementation in Metashape 1.7.0 extensively yet. However, my past experience with the general method of treating scans as spherical panoramas for alignment purposes suggests that the results will likely be very unstable. The resulting panoramas are typically low resolution and, if they are intensity only scans, relatively low contrast; so few tie points are likely to be detected by Metashape's feature detector. From the few small tests I've tried on 1.7.0 that is indeed what seems to be happening.

From a process efficiency perspective, what I want to do as a user is use intensity only scans to capture geometry and photogrammetry to capture color.