What is the recommended settings for high end cameras?

Dear community.

Can someone tell me the recommended settings for processing images from high-end cameras such as the Zenmuse P1 and the Sony RX2?
Processing takes a lot of time, despite having a fairly capable computer:

AMD Ryzen 5950X 16-core processor
128 Gb RAM
RTX 3090 ti

Running WebODM 1.9.16 on Windows 11.

This is the recent attempt with 999 pictures from the Zenmuse P1:

Also, despite having the ignore GSD turned on and the resolution is supposedly 0.88cm ,the orthomosaic is very coarse:

1 Like

Explicitly set your orthophoto-resolution to your target, otherwise it defaults to calculated or 5cm, whichever is greater.


Noted, thanks.

But in this case, hasn’t the program calculated the GSD to be about 0.9cm and is supposed to use that?

1 Like

No. We still default to 5cm. We don’t know what your RAM is, so we are really conservative about settings which might break things for you.

I’m wondering if something like a “Full resolution” preset might make sense for advanced users.

For now, just change your orthophoto-resolution to 0.8, turn off ignore-gsd, and let ODM set it to the calculated. Or since you have decent RAM, make a preset with a resolution of 0.1, keep ignore-gsd off, and all your stuff should come out at its maximum resolution always.


Awesome, thanks! :pray:

Any ideas on how to speed up the processing time for these datasets?
I don’t think I have seen the memory usage go over 32 Gb.
If I select Resize images the processing is much faster, but obviously don’t give the wanted GSD.

Seems like the step that takes the most time is this (17hrs) :
“DEBUG: Matching DJI_20220812125014_0653.JPG and DJI_20220812124357_0282.JPG. Matcher: FLANN (symmetric)”

Here’s a small dataset that I ran, but still rather slow:

1 Like

I thought “ignore GSD” was removed as a settings

1 Like

We have put it back for the time being.

I still want to remove it, or at least make it not as easy to enable but :person_shrugging:

1 Like

Let me clarify recommended settings:

auto-boundary: true, dsm: true, dtm: true, fast-orthophoto: false, ignore-gsd: false, orthophoto-resolution: 0.01

It is odd that matching is taking the longest. This should almost never be true and is usually an indication that some (or all) your data is missing GPS information. At the top of your log, do you have something like what follows?:

[INFO]    ==============
[INFO]    Running dataset stage
[INFO]    Loading dataset from: /var/www/data/4d6304dc-ad69-492c-bc3a-a2714631f2b2/images
[INFO]    Loading 2560 images
[INFO]    Parsing SRS header: EPSG:4326
[INFO]    Updated 2392 image positions
[INFO]    Wrote images database: /var/www/data/4d6304dc-ad69-492c-bc3a-a2714631f2b2/images.json
[INFO]    Found 2560 usable images
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02316_.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02316__.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02316___.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02316____.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02317_.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02317__.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02317___.JPG
[WARNING] GPS position not available for Kakuma_Mission_18A_Flight_02_02318_.JPG

Also, it looks like that dataset ran in less than two hours?

1 Like

Thanks for the recommended settings.

No such message in the console, the images have GPS data.

Yes, the dataset with only 52 images ran in less than two hours.


I ran some images from the same dataset on Lightning.
The processing is fast and the results superb, so it would be interesting to know what the difference in settings is. I used the same settings as in WebODM.



Lightning probably has faster machines than you are running on locally (newer / faster server grade chips and other architecture improvements). That is often an advantage to cloud compute.

1 Like