Agisoft Metashape 1.6.0 pre-release

[Phogi]

The new release looks so good, with several new settings in alignment and empower the processing.

I got two questions about the new release:
(1) I find that camera calibration part set cx,cy as one tick together, and then omitted the p3 p4 parameters, is there any relationship to the new fit additional adjust function? If that is adding p3p4 in, would this add extra error?

(2) if the estimated alignment function should be used after using "source" method and then do not "reset alignment"?

Thanks for answering this!

[Mak11]

With build 9487 I'm getting:

Assertion 239107090 failed at line 7122!

when trying to construct the mesh from depth maps. This occurs both on Ultra and High quality settings.

Hello cbnewham,

This issue will be fixed in the next version update.

Alexey,

Good to know. Just got hit by it when processing on Ultra High finished.

Will the next update also re-enable vertex colors ?

Also, in addition to the low density poly patches randomly scattering the meshes (since the out-of-core mesh update) which I reported https://www.agisoft.com/forum/index.php?topic=11331.msg51488#msg51488 I'm also seeing some other anomalies like seen below:







It happens with main/mesh_tgv_ooc_trimming_enabled True & False.

It would also be good if trimming was dialed a little bit

main/mesh_tgv_ooc_trimming_enabled FALSE



main/mesh_tgv_ooc_trimming_enabled TRUE




Cheers

MAK

[wizprod]

This I would really like to know as well!

Could you please provide more information on the additional calibration coefficients?  What additional coefficients are used compared to the standard Brown model and how are the equations different?

Thank you

I am curious to know what Fit additional corrections option do in Optimize Cameras dialog?....

What are these corrections and when should we use this option?

Hello Paulo,

Fit additional corrections option enables additional calibration coefficients that allows a better fit of the lens distortion model. This is especially important for wide angle lens with large radial distortions, e.g. DJI Phantom 4 RTK camera.

This option is mainly intended to be used in cases when highly accurate reference data is available (RTK/PPK) and it can help to achieve higher accuracy when no GCPs are available.

[Alexey Pasumansky]

Pre-release version has been updated.

Vertex colors calculation has been enabled for the depth maps based mesh (new option in Tools Menu - Colorize Vertices should also allow to apply the colors from the images to the vertices of already generated models). Also the "anomalies" reported by Mak should be fixed (small inverted polygons).

[Mak11]

Alexey,

Awesome work. Will try it asap. Is there any changes to GPU texturing? Are masks and ghosting filter still not supported?

EDIT: Looks like GPU texture generation works with Masks & ghosting filter. Awesome! Just did a quick test. it's was 7x faster than on the CPU. 😮

Cheers

Ike & Tina

[cbnewham]

Select by point is not working how I would expect it to.

Build 9397.

I create the sparse could and see that I have sections which are mis-aligned.
I select some points and press "Filter photos by point". I correctly get a dozen cameras. I reset them and re-align. All OK.

I then see some other points that are not right and select those and follow the same procedure.  In the "details" list of the photos I get a list of several cameras. However, if I Ctr-A them or even select them individually and do "Reset alignment", I am told it's going to reset far more cameras - in fact, I believe the set of cameras to be aligned each time is the summation of previous selections and it is not being reset when a new "filter by point" is done.

In one case I had filtered only 7 cameras (and the list showed only 7 cameras) but the reset said I was resetting 90! If I carry out the realignment it is clearly realigning 90 due to the time it takes.

If I remove filters, select one camera and do "Reset alignment" then it correctly says I am resetting one camera. If I then do select by point as above it all works again as expected.


I wish there was also a way to see what's not aligned clearly in the photo display - something I've requested before. Grey ticks on a dark grey background are next to useless. Perhaps consider allowing us to change the colours for various parts of Metashape? A way to sort in details by "disabled camera" would also be of great assistance.

Thanks - these are small issues in an otherwise fantastic program.

[mrc]

Select by point is not working how I would expect it to.

Build 9397.

I create the sparse could and see that I have sections which are mis-aligned.
I select some points and press "Filter photos by point". I correctly get a dozen cameras. I reset them and re-align. All OK.

I then see some other points that are not right and select those and follow the same procedure.  In the "details" list of the photos I get a list of several cameras. However, if I Ctr-A them or even select them individually and do "Reset alignment", I am told it's going to reset far more cameras - in fact, I believe the set of cameras to be aligned each time is the summation of previous selections and it is not being reset when a new "filter by point" is done.

...

I've had the same bug that has been in every version since 1.3? The problem is that the left hand list of photos retains the previous selection when you use the filter or even just select a range of photos in the "Photos" window.

I work around it by looking for a photo from the filtered list in the left hand list, manually clicking on it in the left hand list, and then performing the mouse drag selection in the filtered list. This always gets the correct photos for realignment.

Note: I did re-report this bug recently so it may be you have a slightly different issue.

https://www.agisoft.com/forum/index.php?topic=11531.0

[Alexey Pasumansky]

Hello cbnewham and mrc,

We'll try to fix it before version 1.6.0 is officially released. It appears that Remove Cameras is working properly from the Photos pane (applied to the selection in the pane only), while Reset Alignment is not considering the pane's selection.

[Phogi]

Hi Alexey,

I'm not sure if this is a bug, when I export depth map using GUI, if I specify the image route individually, it will return error. If I don't specify route but remain it as {filename}, and select folder in the end, it would work.

See attached screen snippings.

Thanks!

[Alexey Pasumansky]

Awesome work. Will try it asap. Is there any changes to GPU texturing? Are masks and ghosting filter still not supported?

EDIT: Looks like GPU texture generation works with Masks & ghosting filter. Awesome! Just did a quick test. it's was 7x faster than on the CPU.

Yes, currently no limitation to masked photos or ghosting filter, also the GPU acceleration is support for the tiled model texture generation sub-step.


The texture atlas size should be, however, a multiple of 4. There are still some other limitations, but mainly for quite uncommon cases related to big number of channels (higher than 6) in the source images.

[James]

Not sure if this is a bug or feature, but using the 'convert images' on a set of fisheye images, the output seems still to be a fisheye rather than rectilinear image.

Is it just that the conversion removes distortions to make a 'perfect' fisheye, or should it be removing the whole fisheye effect plus distortions to make a 'perfect' rectilinear image?

I thought that in previous versions i had been able to use what used to be called 'undistort' to remove the fisheye effect, but that doesn't seem to now be the case, although i can see some subtle differences in the output, still fisheye though!

Images are set to fisheye in 'camera calibration'. There is a single calibration group; just 29 images in the project using nikon d810 and nikon 16mm full frame fisheye, and the reconstruction is very good so i believe the coefficients are right. see attached.

sadly can't share the input images, but if i get a chance will try to recreate the issue with a set safe to share.

[khedar012]

Hey Alexey, thanks for adding support for GPU based Texturing. Are you also planning GPU support for the Densification (not Depth Matching) soon? This also takes really long time for our data sets.

Thanks,
Khedar

[BjFaeTorphins]

I'm getting a crash running build Standard 9617.  There is no error in the log file.  I'm running an align of multiple chunks in batch mode.  It seems to happen during the 2nd chunk.  last entry is...
2019-12-01 15:20:18 Selecting pairs...
2019-12-01 15:20:18 Found 2 GPUs in 0.001 sec (CUDA: 0.001 sec, OpenCL: 0 sec)
2019-12-01 15:20:18 Using device: GeForce GTX 1080 Ti, 28 compute units, free memory: 9233/11264 MB, compute capability 6.1
2019-12-01 15:20:18   driver/runtime CUDA: 10010/8000
2019-12-01 15:20:18   max work group size 1024
2019-12-01 15:20:18   max work item sizes [1024, 1024, 64]
2019-12-01 15:20:18 Using device: GeForce GTX 1080 Ti, 28 compute units, free memory: 9320/11264 MB, compute capability 6.1
2019-12-01 15:20:18   driver/runtime CUDA: 10010/8000
2019-12-01 15:20:18   max work group size 1024
2019-12-01 15:20:18   max work item sizes [1024, 1024, 64]

[Alexey Pasumansky]

Not sure if this is a bug or feature, but using the 'convert images' on a set of fisheye images, the output seems still to be a fisheye rather than rectilinear image.

Hello James,

The undistortion operation result corresponds to the selected camera type, so if the camera type is Fisheye, then you should get an "ideal" fisheye image.

Are you also planning GPU support for the Densification (not Depth Matching) soon? This also takes really long time for our data sets.

Hello Khedar,

Currently we are not planning (at least in short-term future) GPU acceleration of other major workflow steps.

I'm getting a crash running build Standard 9617.  There is no error in the log file.  I'm running an align of multiple chunks in batch mode.  It seems to happen during the 2nd chunk.

Hello BjFaeTorphins,

Was the crash reporter window shown for you?
Also which OS version and which driver version you are using?

[BjFaeTorphins]

I'm getting a crash running build Standard 9617.  There is no error in the log file.  I'm running an align of multiple chunks in batch mode.  It seems to happen during the 2nd chunk.

Hello BjFaeTorphins,

Was the crash reporter window shown for you?
Also which OS version and which driver version you are using?

Metashape exited silently.  no error dialogue or message written to logfile.  Also loses setting to save a logfile and it's location.