WebODM Lightning not processing a 3d image with Jpeg photos converted to Tif using Irfan View

Hi Team

I am not a developer, rather a user of the software.

I am using Opera as a browser, Laptop OS is Windows 11

I have a Parrot Anafi, and am experimenting with creating 3d images of fields in our local area, Christchurch NZ.

Flying at 40m altitude on a flattish field. I have set the a/c to take jpegs at 80% overlap @ camera 70% angle…flights are slow speed.

Using Pix3D to fly a Double Grid flight-plan.

Results in 301 photos taken

I use WebODM Lightning, with credits, to create the 3d image.

Additional Operational parameters of Ortho resolution at 1, pc-quality at high, 3d Mesh true,

Results are good, Avg GSD 3.02cm, Area 34,249m sq, Reconstructed Points 11,709,674

All up 28 minutes to create.

I read(?) that using a TIF photo, gives even better results…

I used Irfan View 64 to convert all the jpegs to TIF, and ran them thru WebODM Lightning with the same operational parmeters as mentioned before.

The project did not work, with a dataset error, see report below. Took 1 hour 30 mins for this message to appear.

Would anyone care to comment: my guess if that the conversion from jpeg to tif isn’t working as required for WebODM to process…… I do not set any additional operational parameters in Irfan for the conversion process……
Message report follows…

" Your task [Task of 2022-02-23T07:41:52.614Z] could not complete. Here’s a copy of the error message:

09:33:22 [App ] Selecting images for dense reconstruction completed: 0 images (518ms)

Fused depth-maps 0 (100%, 0ms)
09:33:22 [App ] Densifying point-cloud completed: 0 points (1s331ms)
09:33:22 [App ] MEMORYINFO: {
09:33:22 [App ] VmPeak: 4004908 kB
09:33:22 [App ] VmSize: 3597044 kB
09:33:22 [App ] } ENDINFO
[INFO] running “/code/SuperBuild/install/bin/OpenMVS/DensifyPointCloud” --filter-point-cloud -1 -i “/var/www/data/371f846a-e601-49a1-802a-51751a715c98/opensfm/undistorted/openmvs/scene_dense.mvs” -v 0 --cuda-device -1
09:33:22 [App ] Build date: Jan 19 2022, 02:40:59
09:33:22 [App ] CPU: AMD Ryzen 9 3900 12-Core Processor (24 cores)
09:33:22 [App ] RAM: 125.90GB Physical Memory 128.00GB Virtual Memory
09:33:22 [App ] OS: Linux 4.15.0-144-generic (x86_64)
09:33:22 [App ] SSE & AVX compatible CPU & OS detected
09:33:22 [App ] Command line: --filter-point-cloud -1 -i /var/www/data/371f846a-e601-49a1-802a-51751a715c98/opensfm/undistorted/openmvs/scene_dense.mvs -v 0 --cuda-device -1
09:33:22 [App ] error: empty initial point-cloud

===== Dumping Info for Geeks (developers need this to fix bugs) =====
Child returned 1
Traceback (most recent call last):
File “/code/stages/odm_app.py”, line 94, in execute
File “/code/opendm/types.py”, line 346, in run
File “/code/opendm/types.py”, line 346, in run
File “/code/opendm/types.py”, line 346, in run
[Previous line repeated 1 more time]
File “/code/opendm/types.py”, line 327, in run
self.process(self.args, outputs)
File “/code/stages/openmvs.py”, line 192, in process
system.run(’"%s" %s’ % (context.omvs_densify_path, ’ '.join(config + gpu_config)))
File “/code/opendm/system.py”, line 106, in run
raise SubprocessException(“Child returned {}”.format(retcode), retcode)
opendm.system.SubprocessException: Child returned 1 "


Hi, I don‘t think that converting will improve your quality.
Since jpeg is the source that will bring, normally no advantage, I guess if you use raw to tiff this could bring more features into the picture since you can have more bits and therefore dynamic.

I guess you refer to ODM 0.9.8 Adds Multispectral, 16bit TIFFs Support and Moar! - OpenDroneMap


As Maurice suggests, converting a JPEG image into TIFF will not give you any improvement in image quality, as details have already been lost in the initial saving as JPEG format. It will make the image files larger though.



Maurice and Gordon have you in the right direction.

You can’t gain information lost by lossy compression, so JPEG to anything else will never help.

However, the Anafi should shoot DNG RAW, which, when converted to TIFF, will have a much higher radiometric resolution and might help with scenes with difficult lighting.

Can you show off a screenshot of what you reconstructed?

Try putting it up in #the-showroom !


Thanx guys

I will look to share the finished result into the showroom

I “understood” that I had saved my files in DNG RAW, but I still need to check this out definitely. The resultant flight photo files still show *.jpg

Copied on converting JPEGs

I’m learning how to get the most bang outta my buck. As well looking to see if I can combine my *.obj results as the area I have scanned has been done several times with the same height and flight characteristics: just different weather. My personal experiment. I understand that its helpful to take several runs at an area with different directions…I guess thats hard on the computing but I do have the time!.


Interestingly several Rugby goal posts do not show, other then base points


Thin linear features often vanish.
Try ultra feature quality and pc-quality, and filter: 5.

Did you resize your images to get GSD=3.2cm, I would have thought it should be <<2cm when flying at 40m.

1 Like

Thanx for the suggestions Gordon
Copied on thin features…would several flights on different lines also assist…or its solely the issue with vertical thinness?
Will try parameters as suggested.
I did not resize images…I recall Pix4Dcaprure said I could get 1.something cm…so, that bit I gotta sort!

1 Like

I suspect the only way to retain thin features - vertical or horizontal is to get some imagery from closer in. I’m yet to test, but perhaps some 70 degree images from an orbit around the posts, from say 20m up might bring them out.
I had a similar issue on a roadside guard rail and posts, which I’m currently trying again with different settings, but suspect re-doing it with closer in images may be the only way to fix it.

1 Like

The filters suggested give me the goal posts…sorta.

But certainly far more detail for me to play with.
Good job, thank you!


You said you’ve got time, right?

Try passing:
--pc-filter 0
--min-num-features 32000

See if that doesn’t get them a bit more :slight_smile:


Will do…
First effort was 28 mins
2nd effort with what you and Gordon suggested was 1hr 30 mins
I’ll revert with time on your additional parameters…for my laptop!


I can see the goal posts well in 3D, additionally the light poles near the Change room building
Good quality result
I have shown a comparison between the parameter changes. with min-num at 32000 & PC-filter 0, less process time


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