Pix4d and Agisoft comparison - high z errors

[Rollo]

I wanted to compare Agisoft Metashape with Pix4d Mapper, so I processed the same data set using both pix4d mapper and the 30 day trial version of agisoft metashape and compared the results. I surveyed an area of 4 hectare, using 6 GCPs and 3 Checkpoints with 50 AGL and a Phantom 4 Advanced. I processed the dataset twice with Agisoft and once with pix4d. The two agisoft datasets only differ in the amount of marked images (I tried to reduce the Error by doing so). The RMSE values can be seen in the attached images. Unfortunately the Agisoft results show a much higher error, especially in the z-Axes, than the pix4d results.
I compared the RMSE of the 3 Checkpoints and the results were (also see images):

Agisoft (few projections) Checkpoint RMSE: 0.008538, 0.008606, 0.111797
Agisoft (lot of projections) Checkpoint RMSE: 0.015079, 0.011417, 0.114200
Pix4D Mapper Checkpoint RMSE: 0.010489, 0.007034, 0.028531


The z-values are pretty high with Agisoft. I also compared the Errors for the GCPs which were (also see images):
Agisoft (few projections) GCP RMSE: 0.018823, 0.011774, 0.066245
Agisoft (lot of projections) GCP RMSE: 0.019157, 0.009321, 0.076240
Pix4D Mapper GCP RMSE: 0.016878, 0.012390, 0.015624
[/b]


 I am wondering if I did something wrong in my workflow and if so, how to improve it. I used SAVs Workflow (https://www.agisoft.com/forum/index.php?topic=7851.0) to process the dataset with Agisoft. To reduce the error furthermore, I tried loads of different things:
Adaptive camera model fitting checked/unchecked
Mark the GCPs and Checkpoints in different images (not blury ones, not over- or underexposed,...)
Check/uncheck Images in Referance Pane for optimization
...

The above mentioned processing options didn´t make much of a change for the RMSE. I would like to learn more about using Agisoft Metashape, but this kind of stopped me from digging deeper into the software. Any help is appreciated!!!

[c-r-o-n-o-s]

These results are not representative. Such a high error can only be due to a wrong setting, sorry...
However, the sources of error are manifold, but you have every time:
"Optimize Cameras" ?

[Alexey Pasumansky]

Hello Rollo,

Can you share the project with the alignment results (re-saved in PSZ format via File Menu -> Save As command) to support@agisoft.com? In case it would be also possible to share the complete image set used for the project, it would be also helpful, as we would be able to check the processing using different settings.

[Rollo]

Hey Alexey, thanks for your reply! I have several options  to save the project as psz file: Metashape archive, metashape 1.6, metashape 1m5. I chose Metashape archive now, I just want to make dute this is the right one. Also do you have a platform/cloud I could upload the images to? It's about 3.26 gb so I can't upload it to wetransfer.

[Alexey Pasumansky]

Hello Rollo,

You can use sharing via Agisoft Cloud free plan account:
https://agisoft.freshdesk.com/support/solutions/folders/31000118608

In this case it is not required to save the project as PSZ, you just need to register and confirm an account on cloud.agisoft.com, enable Cloud processing option in Metashape Network Preferences tab and input the login/password, then use File Menu -> Cloud -> Upload project command. After the synchronization is finished, share the project via personal Agisoft Cloud account page to support@agisoft.com using "Invited people can view and download" option.

[Rollo]

I just read your message! I sent out two mails containing the project and the second one with the images. It is still uploading. Atm I am uploading my project to the cloud, this may take a while. Tonight when I get back home, I can share the link from the cloud with you! Thanks for the awesome and quick support!

[Alexey Pasumansky]

Hello Rollo,

Thank you for sharing the project.

We'll check, what might be a reason of high Z errors observed on some check points. But meanwhile you can enable "fit additional corrections" option in the Optimize cameras dialog (also enable all coefficients), as it should considerably improve the results for the current project.

[Rollo]

Thanks Alexey! I've actually already enabled all coefficients! Do you aalso recommend checking adaptive camera model fitting for a phantom 4 advanced camera? I will try your suggestion tonigjt and enable fit additional corrections.

[Alexey Pasumansky]

Hello Rollo,

You do not need to enable "adaptive camera model fitting".

[Rollo]

Hey Alexey, I wanted to give you an update, your suggestion to enable "Fit additional corrections" and checking all parameters improved the result a lot!!! I attached a screenshot. I am curious, if there are any further improvements I could do to optimize the z error any further. Thanks a lot for your help so far, that´s a great support!

[Rollo]

Also I wanted to ask which settings you would recommend for aligning photos. Should I check both Generic preselection and referenece preselection? Thanks in advance

[Alexey Pasumansky]

Hello Rollo,

For near-vertical (nadir) aerial surveys it is recommended to use Generic + Reference (source) preselection options.

[Rollo]

Thanks a lot Alexey