Two 'surfaces' in point-cloud

How does the quality report look like ?

One or two flights ?

Was a crosshatch flight

The report looks decent although :

• The pixel residual are skewed are bigger than expected.
• Did you flight everything at once, or did the drone landed in between ?
• The GCP residual is what it should be.

But given the two surfaces issues, I’d say it’s more a matching issue. But the residual issue indicates maybe a modeling problem (like we assume one camera model and one shift when maybe with N flight we should consider N).

Do you have/do you know how to find and share with me the config.yaml OpenSfM file ?

Landed, i took 4 batteries

config.yaml OpenSfM file

I imagine it's inside the docker image of nodeodx, i can try to find it

Great, thanks ! Do you have the log by chance ?

It looks a bit like this? https://discord.com/channels/1490218202033029212/1490494474588258436/threads/1526693377473773728 I didn't have time yet to test the proposed solution from piero.

Sure

So I’d say the matching wasn’t powerful enough to bridge some gaps (?) Do you have the stats/stats.json somewhere (it’s the quality report raw data).

Otherwise, happy to play with the data if you can share

Is this related to a split/merge process or something it happen automatically?

When your images don't have enough features / overlap, the program fails to create a single reconstruction; instead, it creates multiple disjoint ones. The program later tries to merge those, but large errors and misalignment can be present.

I see, is there a 'flag' i can use to try to 'force' into one?
Overlap i know it has, maybe increase feature-quality, min-num-features, matcher-neighbors?

Try: Increase min-num-features,
Change feature-type to akaze.
Lower feature-quality to medium (yes, it's counter-intuitive, but it can help reconstruct certain difficult forest areas)

Nice, i queu to my node to run:

1. feature-quality: medium, min-num-features: 60000
2. feature-quality: medium, min-num-features: 60000, feature-type:akaze

In ~12h we will have the results!

Guys, tks a lot for the help

Here are the images, geo.txt and gcp.txt: https://drive.google.com/drive/folders/1SHCEGw2ZTxh4JYIPoJ4ZDEnsHTkDwTYY?usp=drive_link

On both (using akaze and not using it) it still gets the 'two layers'.

Was worth a shot, you probably need more overlap, fly at higher altitude, or both

i will try on a next survey

Tks for the help

Are these flights at different times? Do these flights pause for a battery swap?
I just thought maybe it’s your take off location. If the point you’re taking off between flights is slightly higher or lower it changes your perspective of your camera.
I ran a training flight where I took off from ground first. Swapped a battery and then took off from the top of the case. My gimbal was -45 but my expected agl was off the height of the case; about one foot.
If your mission has locked gimbal angle (obliques) or maybe a locked atl then perhaps a slight change of takeoff height is altering the modeled ground level.

They happen sequencially with pause for battery swap; but take off was from the same location.