WebODM Ignoring Images due to Lighting?

Hi guys, posted this on the Github issues but I was told it’s more of a question for this forum.

WebODM installed with Docker

The issue is that not all uploaded images of the dataset are being used.
I have a dataset of images from F1 - F16, each “F” being a different grid of photos taken/flight, some on different days.
All shots taken on Phantom 3 Standard.
In F6 the photos were taken on an overcast day whereas in F7 it was more of a sunny day so it has more shadows and highlights.
I can generate a model of dataset F1-6 and F7-16 but any model I try to generate using F6 and F7 will only generate up to F6 images or from F7 images onwards.

Map using F7 - F9:

Map using F4 - F9 (Only uses F7 - F9) :

Map using F1 - F14 (Only uses F1 - F6):

I have tried increasing/decreasing image size, only using F6 and F7 (Only uses F7 images), increasing min-num-features.

I think a clue to this issue might be the terrain model, below is an image showing F4-6 and F7-9 generated separately but shown on the same terrain model and the two models aren’t even remotely the same colour.

Did the difference in lighting cause ODM not to be able to find points across the two image sets or did the difference in lighting make ODM calculate height differently across the two and hence not be able to merge them?

Both F6 and F7 were taken at 35 metres using same drone.

------To Reproduce This-----

Any use of F6 and F7 in the same model has caused this issue so it should be easy to replicate.
Link to OneDrive containing F4 - F9 provided, including my LOG when attempting to generate model using F4-F9.

The models won’t be the same color unless you force their stretch to be the same, so I’m not sure the perceived difference is a real difference in elevation.

As far as the batches not matching each other, this is not a surprising outcome – major differences in lighting can definitely affect the ability of the structure from motion process from matching between images (it’s one of the first warnings you get when getting into SfM). If you fly relatively high >120 meters, this effect can be avoided in many cases, but for a 35m flight, matching with such different lighting conditions is unlikely.

1 Like

I have since done multiple more passes and added them images to the overall dataset with the same result but all at 35m.

Do you think a higher flight pass would solve the issue?

A higher pass would likely help, though you may need an updated version of ODM to leverage it.

But fly it and we’ll test.