I’ve processed a data set acquired with a DJI Matrice 4E drone. It was flown with 80/70 overlap and nadir plus sideways oblique images (no forward/backward). Reconstruction detail and accuracy at ground level is excellent, but elements above ground (including a car and a surveillance camera with solar panel) are not reconstructed, even at high detail setting, despite being clearly visible in the tie points:
After some further testing it looks like the default reconstruction region was too low and cut off higher objects. I’ve been able to get them back by extending the reconstruction region.
I’ve also tried various settings to get more detail (detail decimation factor etc.), but they don’t seem to make much difference. Some lead to an increased triangle count, but not the preservation of smaller geometries.
Thank you for the update. That makes sense. If the default reconstruction region was too small, elevated objects would indeed be excluded from the depth map calculation, so extending it was the correct step.
Regarding additional detail: parameters such as the detail decimation factor mainly affect triangle density, not whether small geometries are reconstructed in the first place. An increased triangle count does not necessarily mean better preservation of thin or small objects.
If you want to push details further, the most impactful settings are:
Ensure the reconstruction region comfortably includes all vertical extent
Mesh decimation and detail settings only refine what already exists in the dense cloud; they won’t recover geometry that wasn’t produced during depth-map calculation.
If small elements remain weak, the limiting factors are usually image resolution (GSD) and viewpoint variation rather than mesh settings.
