search

exeter-creww / SFM_Precision

A workflow to create precision maps in python
GNU General Public License v3.0
3 stars 1 forks source link

max precision value #59

Closed h-a-graham closed 4 years ago

h-a-graham commented 4 years ago

So in the most recent merge to the master I have changed the maximum precision value used to fill the gaps in the precision raster.

The reason for this is that one of the precision maps I produced had a very large ~50cm precision value. So I instead switched to using the max precision value in the raster which essentially irons out the extremes.

Are you both okay with this approach? I know we discussed this before and said to use the max precision from the cloud but in hindsight this could be overly cautious where there are weird edge effects for example...

AndrewCunliffe commented 4 years ago

Good question. I just had a chat with Pia and we think that in general, it might be preferable to keep using the maximum point precision for interpolation?

Our reasoning was that this value will be less sensitive to the user-specified grain, so should be more robust across different applications. We also think that where we've got gaps in the sparse cloud, we're expecting the point precision might be worse than the worst precision we do reconstruct, so the conservative approach might be better than smoothing the extremes?

Personally, I'm happier with taking the worst precision value within the area of interest that the resultant LOD maps are clipped to - this might be an alternative / better way of mitigating against edge effects?

piabenaud commented 4 years ago

Yes, great last point there Andy!

h-a-graham commented 4 years ago

Cool, thanks for that both - I'll revert the changes. Andy, any idea how your suggestion could be implemented - can this be extracted from the pdal pipeline?

AndrewCunliffe commented 4 years ago

I think PDAL Info might be the way to go. Implemented on a clipped cloud.

I imagine it should be possible to possible to implement this within the P2G pipeline, although that would mean only rasterising the prescribed AOI - but perhaps that's better as it could also be more efficient?

h-a-graham commented 4 years ago

Okay I'll have a think abou this... it will require a fairly big change to the pipeline as currently the clipping (to an extent) occurs in the gdal writer. Will need to load, clip to extent, clip to aoi (if selected) and then info and export as raster.... Will give this some thought - also not quite sure how to return the info pdal info values?

h-a-graham commented 4 years ago

Right so little update on this. I tried to do the clipping in PDAL (the only way to extract stats on an aoi I think?) but it is crazy slow! So have reverted back to the original plan - precision interpolation is now set to the maximum recorded from the full cloud. I'll close for now but reopen if we think there's a better solution...

AndrewCunliffe commented 4 years ago

Okay sure, I was just looking at the code for this and thinking that it might slow things down quite a bit based on my pipelines...