Point spacing estimation

[pierotofy]

This PR brings in changes to FPCFilter so that we can compute point cloud density estimates.

It also ties in such density estimates into the generation of DEMs, thus disassociating resolution from radius parameters.

This makes it possible to render higher resolution rasters without affecting the radius. It also no longer limits the gsd_scaling factor.

Related #1609 #1608

[pierotofy]

Current master with defaults:

This PR with defaults:

This PR with --dem-resolution 0.000001:

Which looks a bit noisy, but I'd say that's expected given the point cloud density...

[pierotofy]

I'd encourage to test this one as a possible solution to #1608 @smathermather and @Saijin-Naib, in particular the method for point cloud density estimation, which I opted to do in a somewhat novel way but has not been tested. (Check if the density value in odm_filterpoints/point_cloud_stats.json makes sense if something weird happens).

[smathermather]

I probably won't have time to test as I'm traveling this week, but @Saijin-Naib -- there will be a node at 3002 (already added and permissions shared with you) if you've got capacity to throw some dataset through.

It's not ready yet, but should be within the hour (it's building).

[Saijin-Naib]

Yeah, absolutely, I'll get on that throughout the week

[pierotofy]

A lot of changes, I'm noticing improvements to orthophotos across the board as well, but more testing is needed before I feel confident to merge.

[Saijin-Naib]

This makes me happy!

I have all of our bolded ODMData datasets enqueued for processing, so we have decent coverage coming.

Any non-bolded anyone wants to see?

[pierotofy]

sheffield_park_1, sand_key would be interesting.

[pierotofy]

Biggest change should be a noticeable decrease in average relief displacement around buildings:

Master:

This PR:

[Saijin-Naib]

This is incredibly exciting, as I get this as a point of feedback quite consistently.

Extrapolating from here, I'd expect to improve reconstruction on solar panels as well, but we've not had any datasets submitted for that type of scenery, unfortunately.

[Saijin-Naib]

I'm still collecting test results, but at first blush, your changes have massively improved the appearance of tree canopies, which is always nice to note.

[Saijin-Naib]

Quick Comparison with some key areas of difference on mf_old_orchard (4.56 cm/px GSD):

crop: 0, dem-resolution: 0,1, dsm: true, dtm: true, feature-quality: ultra, orthophoto-resolution: 0.1, pc-filter: 0, pc-quality: ultra

Some quick things that seem to have improved:

• Tree Canopies
• Straight Lines (crop rows, traffic markers, building edges, etc)
• Relief/Extrusion (individual drainage tiles are visible)

It seems almost like what currently is Ultra ends up being more like Medium or Low in comparison to experimental v3.0.5 4a0b60b.

Rest of the datasets are processing up still.

[atobiese]

Great work, thanks to all of you for this. I will run testing with this last commit. I tried to share the discussion regarding the scaling factor from the forum to @pierotofy, but does not seem like odm-discourse is set up for sharing dm's. We discussed the improvements in dem's with removal of the gsd cap compare to that of the orthophoto, use case is DEM over snow. Looking forward to testing and will get back with results.