Agisoft Metashape 1.7.0 pre-release

[JRM]

Alexey > could you please give us more info about connected components and their meaning ? It is difficult to try out on a feature which has no clear use case

[SB]

[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.

...

Yes, it can be said that the laser scans are aligned together with the common images.

If you need to rely on the accurate scanner locations, you can input corresponding coordinates to the Reference pane for the related camera instances and adjust the accuracy value for them in order to increase the weight for them.

...

I would also like to see an option to just check a box or similar to force laser scan alignment to be preserved exactly, if it had been registered already in a 3rd party package, otherwise the function would not be useful. It seems a lot of manual work to input the coordinates of each laser setup (and presumably orientation?) to ensure they stay fixed. I take great care over my registration and wouldn't trust any algorithm to do a better job! We treat our laser data as sacred and once it is registered everything else (so many things...) is derived from it, but can not be allowed to affect it!

I haven't actually tried 1.7 yet so I might not know what I'm talking about, but it seemed like a potential issue!

I would also like for the laser scan data to remain registered.  I am trying it now for the past few days and making some progress.

I had to use 40 markers to align 4 laser scans and 4 360 degree panorama photos due to the spacing between the laser scans and also no transition from camera positions.  However, the laser scan data came directly from FARO Scene ready to go.

The imported laser data does have reference metadata such as position, yaw, pitch, roll but when I try to use that information and align by reference, the program crashes and sends an error report.  I imagine that eventually, the scan data can be positioned using the reference metadata and forced to stay in that position by placing markers and assigning xyz values but it sure would be nice to have that done by default.

Thank you for a great start!

[Paulo]

I got following error generating Dense cloud:

2020-11-07 00:06:53 BuildDenseCloud: quality = High, depth filtering = Mild, PM version, point colors = 1
2020-11-07 00:06:53 Generating depth maps...
2020-11-07 00:06:53 Preparing 48 cameras info...
2020-11-07 00:06:53 Finished processing in 0.026 sec (exit code 0)
2020-11-07 00:06:53 Error: Assertion 23910910010 failed at line 54!

The data set consists of 48 Phantom3 images with rolling shutter compensation on. Optimization was done with Fit Additional parameters on...

[laurent.G]

I ve try to install pre-release on RYZEN 3900x Windows 10 but installer finish without install metashape.
What can i do ?

Hello Laurent,

Are you observing any errors during the installation? If there are any, please provide the related screenshots or specify exact error message observed.

hello Alexey
i don't have a special error only this message...
thanks

hi alexey,
i 've sucessefully install , i've stop real time virus protection of win 10

thanks

[OPK Solutions]

I got following error generating Dense cloud:

2020-11-07 00:06:53 BuildDenseCloud: quality = High, depth filtering = Mild, PM version, point colors = 1
2020-11-07 00:06:53 Generating depth maps...
2020-11-07 00:06:53 Preparing 48 cameras info...
2020-11-07 00:06:53 Finished processing in 0.026 sec (exit code 0)
2020-11-07 00:06:53 Error: Assertion 23910910010 failed at line 54!

The data set consists of 48 Phantom3 images with rolling shutter compensation on. Optimization was done with Fit Additional parameters on...

I Reported this as a bug some days ago: once you switch the compensation off, some tears of pain are shed due to high pixel matching errors, but still you get your point cloud. At this stage I'm wondering if this compensation is of any use as there is the option for Image-Variant Parameters. With the new pre-release version I can't even align images, but at least can see that the same rolling shutter compensation issue hasn't been completely addressed yet on projects aligned using the previous pre-release. Still, the software has included some nice settings, and is creating better models than previous official releases, I find.

Os

[Diego]

Rolling shutter compensation on. At this stage I'm wondering if this compensation is of any use as there is the option for Image-Variant Parameters.

Rolling shutter compensation; Of course it works and is of great help, and has nothing to do with Image-Variant Parameters, this version is not final yet, if you want to use rolling shutter use version 1.6.5 in the meantime.

[Paulo]

 I experience many Metashape crashes when aligning 527 images from DJI Phantom 4 Pro. It exits from Metashape with crash report window appearing. It seems to happen during Selecting pairs ... just after KeyPoint detection..... See attached log. Also attached is image of GPU performance monitor just after crash (indicated by red vertical line).....

I also experienced same crash trying to align a set of Mavic 2 Pro images.....

Update: got same crash with a set of 101 images from SODA camera EbeeX

[Daveblueband]

[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.

...

Yes, it can be said that the laser scans are aligned together with the common images.

If you need to rely on the accurate scanner locations, you can input corresponding coordinates to the Reference pane for the related camera instances and adjust the accuracy value for them in order to increase the weight for them.

...

I would also like to see an option to just check a box or similar to force laser scan alignment to be preserved exactly, if it had been registered already in a 3rd party package, otherwise the function would not be useful. It seems a lot of manual work to input the coordinates of each laser setup (and presumably orientation?) to ensure they stay fixed. I take great care over my registration and wouldn't trust any algorithm to do a better job! We treat our laser data as sacred and once it is registered everything else (so many things...) is derived from it, but can not be allowed to affect it!

I haven't actually tried 1.7 yet so I might not know what I'm talking about, but it seemed like a potential issue!

I would also like for the laser scan data to remain registered.  I am trying it now for the past few days and making some progress.

I had to use 40 markers to align 4 laser scans and 4 360 degree panorama photos due to the spacing between the laser scans and also no transition from camera positions.  However, the laser scan data came directly from FARO Scene ready to go.

The imported laser data does have reference metadata such as position, yaw, pitch, roll but when I try to use that information and align by reference, the program crashes and sends an error report.  I imagine that eventually, the scan data can be positioned using the reference metadata and forced to stay in that position by placing markers and assigning xyz values but it sure would be nice to have that done by default.

Thank you for a great start!

Hello, I also would recommend lidar being considered registered on import as well. I’ve been currently trying to setup FARO Scene data being sent into Metashape in the past (I do not have the ability to demo 1.7 at this time with work). Often I need to generate textures/missing areas on already registered and locked lidar data so having this option would be helpful.

I am curious as to how points/normal generate is handled - is there by chance documentation outlining how these work together? I’m currently assuming the lidar is converted into a six pack camera station image of the lidar with tie points to it’s own lidar cloud data, but if there are any videos/documentation outlining this more, it would help in providing some feedback.

Also, is it possible to transfer point normals to another lidar cloud import (i.e. I have a basic setup of the lidar registered and I have a cleaned evenly dense cloud that I want to transfer/interpolate point normals from one to the other so as to mesh that cloud instead). If this doesn’t exist with the 1.7 demo, I’ll see about putting in a feature request.

Thanks!

[wojtek]

Hey,

Can we also have a Total Points argument in the python implementation of SamplePoints? (if we want exact amount of points).

Edit: on some testing I also got a following faulty result (see image)

[Alexey Pasumansky]

I experience many Metashape crashes when aligning 527 images from DJI Phantom 4 Pro. It exits from Metashape with crash report window appearing. It seems to happen during Selecting pairs ... just after KeyPoint detection..... See attached log. Also attached is image of GPU performance monitor just after crash (indicated by red vertical line).....

I also experienced same crash trying to align a set of Mavic 2 Pro images.....

Update: got same crash with a set of 101 images from SODA camera EbeeX

Hello Paul,

There was a bug related to the image matching stage, when the incremental image alignment feature is used (keep points stored in the project and new images are added), it should be fixed in the next pre-release update. But can you please specify, whether you are observing the similar issues even for newly created project with the initial image matching operation?

[Paulo]

Hello Alexey,

I started from scratch, loading 101 images from SODA camera and started image alignment and got same crash after image key point detection and starting Selecting pairs .... The crash report was sent....At moment of crash both Compute_0 and 3D processes of GPU spike to 100%.
Here is log collected til crash:

Code: [Select]

2020-11-21 11:15:36 Loading photos...
2020-11-21 11:15:38 Finished processing in 2.586 sec (exit code 1)
2020-11-21 11:16:46 AlignPhotos: accuracy = High, preselection = generic, reference, keypoint limit = 50000, tiepoint limit = 10000, apply masks = 0, filter tie points = 0, adaptive fitting = 1
2020-11-21 11:16:46 Matching photos...
2020-11-21 11:16:46 Detecting points...
2020-11-21 11:16:46 Found 1 GPUs in 0.002 sec (CUDA: 0.002 sec, OpenCL: 0 sec)
2020-11-21 11:16:49 Using device: GeForce GTX 670MX, 5 compute units, free memory: 2494/3072 MB, compute capability 3.0
2020-11-21 11:16:49   driver/runtime CUDA: 10010/8000
2020-11-21 11:16:49   max work group size 1024
2020-11-21 11:16:49   max work item sizes [1024, 1024, 64]
2020-11-21 11:16:49   got device properties in 0.001 sec, free memory in 2.893 sec
2020-11-21 11:16:50 [GPU] photo 1: 50000 points
2020-11-21 11:16:52 [GPU] photo 2: 50000 points
2020-11-21 11:16:52 [CPU] photo 0: 50000 points
2020-11-21 11:16:53 [GPU] photo 3: 50000 points
2020-11-21 11:16:55 [GPU] photo 5: 50000 points
2020-11-21 11:16:56 [GPU] photo 6: 50000 points
2020-11-21 11:16:56 [CPU] photo 4: 50000 points
2020-11-21 11:16:57 [GPU] photo 7: 50000 points
2020-11-21 11:16:59 [GPU] photo 9: 50000 points
2020-11-21 11:17:00 [CPU] photo 8: 50000 points
2020-11-21 11:17:00 [GPU] photo 10: 50000 points
2020-11-21 11:17:01 [GPU] photo 12: 50000 points
2020-11-21 11:17:03 [GPU] photo 13: 50000 points
2020-11-21 11:17:04 [CPU] photo 11: 50000 points
2020-11-21 11:17:04 [GPU] photo 14: 50000 points
2020-11-21 11:17:05 [GPU] photo 16: 50000 points
2020-11-21 11:17:07 [GPU] photo 17: 50000 points
2020-11-21 11:17:07 [CPU] photo 15: 50000 points
2020-11-21 11:17:08 [GPU] photo 18: 50000 points
2020-11-21 11:17:09 [GPU] photo 20: 50000 points
2020-11-21 11:17:11 [GPU] photo 21: 50000 points
2020-11-21 11:17:11 [CPU] photo 19: 50000 points
2020-11-21 11:17:12 [GPU] photo 22: 50000 points
2020-11-21 11:17:13 [GPU] photo 24: 50000 points
2020-11-21 11:17:15 [CPU] photo 23: 50000 points
2020-11-21 11:17:15 [GPU] photo 25: 50000 points
2020-11-21 11:17:16 [GPU] photo 27: 50000 points
2020-11-21 11:17:17 [GPU] photo 28: 50000 points
2020-11-21 11:17:18 [CPU] photo 26: 50000 points
2020-11-21 11:17:19 [GPU] photo 29: 50000 points
2020-11-21 11:17:20 [GPU] photo 31: 50000 points
2020-11-21 11:17:21 [GPU] photo 32: 50000 points
2020-11-21 11:17:22 [CPU] photo 30: 50000 points
2020-11-21 11:17:23 [GPU] photo 33: 50000 points
2020-11-21 11:17:24 [GPU] photo 35: 50000 points
2020-11-21 11:17:25 [GPU] photo 36: 50000 points
2020-11-21 11:17:26 [CPU] photo 34: 50000 points
2020-11-21 11:17:27 [GPU] photo 37: 50000 points
2020-11-21 11:17:28 [GPU] photo 39: 50000 points
2020-11-21 11:17:29 [CPU] photo 38: 50000 points
2020-11-21 11:17:29 [GPU] photo 40: 50000 points
2020-11-21 11:17:31 [GPU] photo 42: 50000 points
2020-11-21 11:17:32 [GPU] photo 43: 50000 points
2020-11-21 11:17:33 [CPU] photo 41: 50000 points
2020-11-21 11:17:33 [GPU] photo 44: 50000 points
2020-11-21 11:17:35 [GPU] photo 46: 50000 points
2020-11-21 11:17:36 [GPU] photo 47: 50000 points
2020-11-21 11:17:37 [CPU] photo 45: 50000 points
2020-11-21 11:17:37 [GPU] photo 48: 50000 points
2020-11-21 11:17:39 [GPU] photo 50: 50000 points
2020-11-21 11:17:40 [GPU] photo 51: 50000 points
2020-11-21 11:17:41 [CPU] photo 49: 50000 points
2020-11-21 11:17:41 [GPU] photo 52: 50000 points
2020-11-21 11:17:43 [GPU] photo 54: 50000 points
2020-11-21 11:17:44 [GPU] photo 55: 50000 points
2020-11-21 11:17:45 [CPU] photo 53: 50000 points
2020-11-21 11:17:45 [GPU] photo 56: 50000 points
2020-11-21 11:17:47 [GPU] photo 58: 50000 points
2020-11-21 11:17:48 [CPU] photo 57: 50000 points
2020-11-21 11:17:48 [GPU] photo 59: 50000 points
2020-11-21 11:17:50 [GPU] photo 61: 50000 points
2020-11-21 11:17:51 [GPU] photo 62: 50000 points
2020-11-21 11:17:52 [GPU] photo 63: 50000 points
2020-11-21 11:17:53 [CPU] photo 60: 50000 points
2020-11-21 11:17:54 [GPU] photo 64: 50000 points
2020-11-21 11:17:55 [GPU] photo 66: 50000 points
2020-11-21 11:17:57 [GPU] photo 67: 50000 points
2020-11-21 11:17:57 [CPU] photo 65: 50000 points
2020-11-21 11:17:58 [GPU] photo 68: 50000 points
2020-11-21 11:17:59 [GPU] photo 70: 50000 points
2020-11-21 11:18:01 [CPU] photo 69: 50000 points
2020-11-21 11:18:01 [GPU] photo 71: 50000 points
2020-11-21 11:18:02 [GPU] photo 73: 50000 points
2020-11-21 11:18:03 [GPU] photo 74: 50000 points
2020-11-21 11:18:04 [CPU] photo 72: 50000 points
2020-11-21 11:18:05 [GPU] photo 75: 50000 points
2020-11-21 11:18:06 [GPU] photo 77: 50000 points
2020-11-21 11:18:08 [GPU] photo 78: 50000 points
2020-11-21 11:18:08 [CPU] photo 76: 50000 points
2020-11-21 11:18:09 [GPU] photo 79: 50000 points
2020-11-21 11:18:10 [GPU] photo 81: 50000 points
2020-11-21 11:18:12 [GPU] photo 82: 50000 points
2020-11-21 11:18:12 [CPU] photo 80: 50000 points
2020-11-21 11:18:13 [GPU] photo 83: 50000 points
2020-11-21 11:18:14 [GPU] photo 85: 50000 points
2020-11-21 11:18:16 [CPU] photo 84: 50000 points
2020-11-21 11:18:16 [GPU] photo 86: 50000 points
2020-11-21 11:18:17 [GPU] photo 88: 50000 points
2020-11-21 11:18:18 [GPU] photo 89: 50000 points
2020-11-21 11:18:19 [CPU] photo 87: 50000 points
2020-11-21 11:18:20 [GPU] photo 90: 50000 points
2020-11-21 11:18:21 [GPU] photo 92: 50000 points
2020-11-21 11:18:23 [GPU] photo 93: 50000 points
2020-11-21 11:18:23 [CPU] photo 91: 50000 points
2020-11-21 11:18:24 [GPU] photo 94: 50000 points
2020-11-21 11:18:25 [GPU] photo 96: 50000 points
2020-11-21 11:18:27 [GPU] photo 97: 50000 points
2020-11-21 11:18:27 [CPU] photo 95: 50000 points
2020-11-21 11:18:28 [GPU] photo 98: 50000 points
2020-11-21 11:18:30 [GPU] photo 100: 50000 points
2020-11-21 11:18:31 [CPU] photo 99: 50000 points
2020-11-21 11:18:31 points detected in 105.068 sec
2020-11-21 11:18:31 Selecting pairs...
2020-11-21 11:18:31 Found 1 GPUs in 0.001 sec (CUDA: 0.001 sec, OpenCL: 0 sec)
2020-11-21 11:18:31 Using device: GeForce GTX 670MX, 5 compute units, free memory: 2503/3072 MB, compute capability 3.0
2020-11-21 11:18:31   driver/runtime CUDA: 10010/8000
2020-11-21 11:18:31   max work group size 1024
2020-11-21 11:18:31   max work item sizes [1024, 1024, 64]

Note hat Keep keypoints option was not checked.....

[Alexey Pasumansky]

Hello Paul,

Thank you for additional information.

I think you should be able to avoid these crashed in the current pre-release by disabling CPU utilization in the GPU Preferences tab. But the proper fix will be implemented in the next 1.7.0 update.

[Paulo]

Thanks Alexey,

I was able to do alignment on this 101 image data set disabling the  Use CPU in the GPU preferences. However Alignment Time is much greater compared to 1.6.5. See attached file.

Note that Image matching increases dramatically from 16 min to 55 min while Image alignment time reduces a little bit from 5 min to under 4 min.

I do hope  image matching time will be better optimized in future release!

[Alexey Pasumansky]

Hello Paul,

I have used the banchmarking dataset in the versions 1.6.5 and 1.7.0 and got comparable timing for image matching:
https://www.dropbox.com/s/hh5yg0fmpr4bpn3/benchmarking_1.3.zip?dl=0
The same processing settings have been used (similar to your test) and also fine level task subdivison has been disabled (the processing performed in Untitled - unsaved project).

Do you see in the logs that GPU is used in both runs properly, without any fallbacks?

[Paulo]

Hi Alexey,

I think that the difference is due that for lower mid level cards as Geforce GTX 670MX using CPU in GPU processes does affect greatly performance. In previous test for 1.6.5. the Use CPU was checked in GPU options. I redid test without checking Use CPU and get following results:

Alignment parameters   
Accuracy   High
Generic preselection   Yes
Reference preselection   Source
Key point limit   50,000
Tie point limit   10,000
Guided image matching   No
Adaptive camera model fitting   Yes
Matching time   58 minutes 37 seconds
Matching memory usage   185.48 MB
Alignment time   3 minutes 53 seconds
Alignment memory usage   66.07 MB
Software version      1.6.5.11249

The 58 min of matching time is comparable to the 55 min obtained in version 1.7.

So it is hoped that with next release, Using CPU in GPU processes will also considerably reduce Matching time.....