Best River mapping settings for webODM and comparison to DD


I’ve been experimenting with a few different settings on webODM for mapping both complex evergreen forests and rivers, with some decent results. But I’m yet to find a set of settings which don’t have warping issues or mismatches. Comparing to DD, WebODM does a good job but seems to show some inconsistencies no matter what settings I try. Flying with a 75% front and side overlap at 150m.

My biggest issue is that with webODM, vegetation seems to be stitched into areas of river no matter what settings I try. Any ideas on whether I can get a better set of webODM settings for a similar output to DD (will post in comments as I am not able to add more than one embedded image as a new user)

WebODM - dem-resolution: 2.0, dsm: true, gps-accuracy: 1, min-num-features: 15000, orthophoto-resolution: 2.0, pc-quality: high, use-3dmesh: true

1 Like

DD output for comparison - note that the DD output shows all areas of river and back channel and doesn’t have the same gaps/warps in the ortho.

1 Like


Are you on the latest releases of WebODM?

Have you experimented with enabling rolling shutter distortion correction? What sUAS are you using?

1 Like

Thanks Saijin

  • Using a M2P, with the latest release of WebODM on Mac OS through Docker.

I’ve tried using fixed camera parameters with the following output - slightly better output in some areas but still some issues with capturing the whole river.

Settings - dem-resolution: 2.0, dsm: true, matcher-neighbors: 16, min-num-features: 12000, orthophoto-resolution: 2.0, pc-quality: high, use-3dmesh: true, use-fixed-camera-params: true, rerun-from: dataset


I wouldn’t recommend using fixed camera parameters.

Can you bump feature-quality to high or ultra, and try enabling rolling shutter correction?


Still getting similar quality with rolling shutter on and features set to high. Ultra doesn’t appear to work on my computer due to memory usage. There is one particular area which has overlap of trees with the river which I can’t seem to correct and I don’t get similar issues in Dronedeploy.

d-tiles: true, auto-boundary: true, gps-accuracy: 0.2, matcher-neighbors: 50, mesh-size: 30000, min-num-features: 18000, orthophoto-resolution: 1, pc-geometric: true, pc-quality: high, rolling-shutter: true, use-3dmesh: true


Can you try adding crop 0 as well?

1 Like

Hi ctorm,

It’d be helpful if you screenshot and show side-by-side your target appearance and the issues you’re seeing. It is hard to tell much from screenshots of the whole area.

You can check out this version:

auto-boundary: true, dsm: true, feature-quality: ultra, pc-quality: ultra, rolling-shutter: true, rolling-shutter-readout: 56, 

I also recommend trying the “fields” drop down setting. Your sidelap is pretty low for a forest area, so true orthos can be a challenge, but planar estimates (which is likely what DD if failing back to here) may look better.


Here’s what it looks like for me using ultra/ultra:
63 images 02:34:35
||auto-boundary: true, dem-resolution: 2, dsm: true, feature-quality: ultra, gps-accuracy: 6, mesh-size: 250000, min-num-features: 15000, orthophoto-resolution: 1, pc-quality: ultra, resize-to: -1, use-3dmesh: true|
|Average GSD:|3.53 cm|
|Area:|197,671.64 m²|
|Reconstructed Points:|23,539,107|

in 3D


Thanks for this smatehrmather. I think you may have solved it with the planar estimates, I’ll do a re-run now but from your version it looks like most of the issues have been corrected. I’ll upload a side-by side of any areas I have issues as I understand it’s difficult to compare - I couldn’t do that initially as a first time poster.

Good to know about the planar estimates - it can be difficult to get a flight with good (over 90%) overlap without massive battery use in some of the remote forest settings I map and having used DD for quite a while now I’m guessing that’s what they revert to with their recommended flight path settings.


1 Like

Yeah, forests are tricky. I have found with recent improvements to OpenDroneMap, cranking the quality settings is a good if not perfect substitute for 80+% overlap. I flew a forest in Vermont this week and forgot my charger, so I only got half the images I meant to, and by turning up feature-quality and pc-quality got a quite adequate result.

Planar (fields) may work well here too. I haven’t tried it with your dataset, but it is a much less memory intensive approach to try.

1 Like

I think you need to set the sensor readout speed as I don’t think M2P is in the database for auto rolling shutter removal, at least it wasn’t when I tried a couple of days ago.

1 Like

Correct. If you use these settings, you should be pretty happy, though the ram in your machine might not be:

auto-boundary: true, dsm: true, feature-quality: ultra, pc-quality: ultra, rolling-shutter: true, rolling-shutter-readout: 56, 
1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.