Possible bug in orthophoto meshing in 2.1.0

Interesting possible bug in meshing for the orthophoto. This is a dataset that previously worked in 1.x series (untested in 2.0.0):

The point cloud looks good:
image

The 3D mesh looks good:
image

The DSM is flipping spectacular:
image

But the ortho is a mess:
image

Console text can be found here:

Images can be shared as needed.

1 Like

Mm, interesting!

Images would help, as well as testing on 2.0.0 and see if it happens there.

You can pull the 2.0.0 vesion with docker pull opendronemap/nodeodm:2.0.0.

2 Likes

Images are (or soon will be) here:

I can run it through a 2.0.0 node shortly.

1 Like

No access :confused: link asks for a login.

Stupid OneDrive. It looks like you have to have an MS account to access data shared with the world. Try this:

It’s uploading now, so should be ready 9:45 ET

1 Like

Ah, that’s more an Organizational restriction. No such limitation for data hosted in Personal OneDrive accounts.

Yell at your sys-admin (nicely :wink: ).

Ah, that must have been part of the recent security lock-down. Let’s see if we can get them to ease that a bit: I understand not having write access, but if I want to share things widely, I need that capacity.

1 Like

Maybe pitch it to them with Time-Based Expiry sharing as well, so the link doesn’t stay active indefinitely. In addition, a PSK (pre-shared key) can be added to further secure access to just those whom have the password you assign to the share link.

It shouldn’t present too much of a risk surface for them, I hope!

1 Like

Yeah, I think I can pitch time-based expiry without too much fuss.

1 Like

Mm, I got a nice looking ortho using default settings:

image

Are you on Windows and did you mount a volume by any chance when running ODM?

1 Like

Interesting. My parameters weren’t too strange:
Options: mesh-octree-depth: 10, texturing-nadir-weight: 28, mesh-size: 300000, dsm: true, camera-lens: brown, depthmap-resolution: 1000

Sorry, didn’t see the Windows question.Nope here too. :smiley:

I’ll try rerunning.

Huh. Rerunning resulted in a good ortho. Super weird. I’ll close this and monitor… .

2 Likes

Hmm, another case of it with the Aukerman dataset:
image

I have a feeling if I rerun It will be fine.

Do the point cloud / textured models look fine?

Likely, this looks like a corruption / race condition issue.

1 Like

An interesting thing to try would be to run the pipeline from the mvs-texturing step multiple times (say 10 or 20) and keep copies of the resulting orthophotos, see how many of those are corrupted.

1 Like