search

open-forest-observatory / automate-metashape

Easy, reproducible Metashape photogrammetry workflows
Other
76 stars 27 forks source link

Erroneous height-above-ground estimate leads to low resolution DEM #62

Open russelldj opened 2 months ago

russelldj commented 2 months ago

I was trying to run reconstruction on what appears to be a fairly-typical drone survey provided by a collaborator. Even though the data appeared to be captured at a moderate altitude (e.g. 50-100m) metashape reported that it was 410m. This caused the resolution of the DEM and resultant orthomosaic to be very low. I'm running some additional experiments to try to diagnose the problem but wanted to avoid this getting lost in the mean time.

I'm happy to share the data internally as needed.

russelldj commented 2 months ago

Even an orthomosaic built from the mesh has the same low resolution.

russelldj commented 2 months ago

I'm now much more confused than before. I reran with the dem resolution (here) parameter set to 5cm, which in turn would force a 1.25cm/pix ortho derived from the mesh-based DSM. The result had the expected resolution, but the result was nonsense. It consisted of a patchwork of rectangles, which each looked like they contained a zoomed in version of the image chip. Another things which seems odd is in the pdf report, the surveyed region looks very small compared to the camera, like the rendered view is way off. This has happened across all runs. I have completed one full run from scratch to ensure it wasn't a one-off error.

Again, happy to share internally and I'll reach out to person who provided the data to see if they have any insights. Nothing about the data seems out of the ordinary.

russelldj commented 2 months ago

Somehow I missed this earlier. The collaborator shared an orthomosaic that they produced. It seems completely fine and has sub-cm resolution. I assume it was produced using the GUI.

youngdjn commented 2 months ago

This could be revealing a cryptic bug in automate-metashape-2 that only appears given certain qualities in the input data (and has not occurred with our own datasets). Can you confirm that you can process this external dataset properly in the GUI? That will help confirm whether it is an automate-metashape-2 bug or something more general.