Processing time compare WebODM vs. Pix4DFields

Hi folks,

My topic is about to understand how far the processing time is so different between WebODM and Pix4DFields, assuming the same CPU, memory, same machine but using Docker for WebODM.

For example, the same dataset used to compare is 1600 photos taken using Micasense Altum

Pix4DFields (Less 1 hour)
*Generate ortho (full size) w. reflectance panel correction
*Generate indexes NDVI, VARI, NDRE

WebODM (~9 hours)

  • Generate ortho and algo (indexes) without resizing

In both cases we use a Windows i7 10750H, GeForce RTX2070 8 Go, 16Go. Docker desktop for WebODM.

I look to be able to use WebODM or ODM (not played yet) to achieve the similar amount of time.

How is possible ?

1 Like

Pix4D Fields is doing a 2D-only reconstruction.

You can get a more similar processing pipeline by using the Fast Orthophoto option.

The Fast Orthophoto option will also take care of my multispectral images (6 bands) ?

1 Like

No, you’ll need to modify it to include the --radiometric-calibration option.

I’m going to try it right now and let you know. Thanks

1 Like

I made some tests today to compare. Currently I run a processing of 1512 images and I have uncheck some option like 3D model to save time. What’s is wrong in the workflow ?

After 3h and 12 minutes my processing is done and even if I checked Skip 3D model, the 3d model was generated. Its possible my options selection is not working ?


I do many test to processing and even I check skip-3dmodel:true or for example select the processing fast-ortho, the 3D Model is always generated and consume a lot of time to proceed. You can look my both screenshots and see the “Options”.

What’s wrong ?

–skip-3dmodel skips the generation of a true 3D model, not the creation of a more simplified 2.5D extruded model.

Currently, our processing pipeline will still reconstruct that at the least

If I looking to reduce the processing time and I don’t need 3D model, just ortho and plant health with radio-correction: camera-sun which options I have to use ?

Photos : 480
Options: radiometric-calibration: camera, skip-3dmodel: true, texturing-skip-global-seam-leveling: true
Duration : 02:10:59

I ask because I run many processing tests and I not able to see any improvements ?

Note: My post concern using Micasense Altum camera

You could try --auto-boundary, --feature-type ORB maybe. If you’re using ORB, you can safely increase the --min-num-features to more than 10x what they currently are without issue as it runs fast/

@Saijin_Naib I have made some tests with your recommendation and I am not able to have an accurate processing time for our workflow. Again the processing time is around 3 hours. I understand for someone here 3 hours is fast but in our workflow WebODM with 3 hours processing time over ~40 minutes with Pix4DFields is too much time.

Maybe I have missed a guide to use WedODM correclty ?

16GB RAM is pretty tight. Are you swapping out to the pagefile during reconstruction? If so, things are going to get a bit slow during parts of the reconstruction pipeline.

Also, did you use ORB as I recommended? It should make a drastic difference during the matching phase, and reliably reduces dataset processing times here by about 5x or more.

Was 32GB RAM not 16GB (was my information error)

Yes I used ORB and I see the time reducing from 3.5 hours down to ~1.5 hours if I remember

1 Like

Hey, about a half hour difference from Pix4D Fields! I’ll take that!

You may be able to reduce matcher neighbors a bit if overlap/sidelap are good enough and ORB is finding features well enough.

That can save you lots of time, too.

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