Tony, this looks quite good ... and I can definitely understand the problems involved with your approach. It is - simply put - the same issue, which scientists face when trying to automatically classify (land cover / land use) sat / aerial imagery into landclass data. And usually its almost impossible to do it right by simply looking at RGB (true color) imagery ... thats why pro classification approaches all use multispectral (more than just RGB channels) data

By the way ... for the USA, I think you might - instead of doing it algorithmically yourself (with all the pitfalls included) - really think about falling back on NLCD data for example, which is really detailed and should be aligned quite well with most of the orthoimagery (is also very high res). It also already gives you more than just forest, non-forest info ... but also if you have shrub / conifers / broadleaved (and if you factor in NLCD canopy data, you can even differentiate between sparse / dense forests). The default and HD Mesh use this data too ... but based on the generated triangle mesh resolution do not fully capture the max possible detail included in the data.

My first X-Plane project for XP8 (where the default scenery did not even have any forests) and XP9 (where the default forests were extremely coarse) was about generating forest overlays based on NLCD (for the US) and CORINE (for Europe).

1 hour ago, wiler said:

Ortho is a game changer, yes there are trees in wrong locations still but by in large its a major step forward in mesh and can be insanely realistic. 

I hate to be a constant Negative Nellie on this subject, but once in a while I think it's worth mentioning that we don't all love the orthophoto concept. And there is that topic header up there, so...

I just can't get past the artifacts at low flight levels -- the glaring contrast between razor-sharp detailed 3D autogen buildings and the blurry ground they sit on. Just two quick examples below, from that Helsinki clip posted above. The agricultural fields look much better than generic landclass. But then I see something like that squashed athletic field at the left (because there is no equivalent 3D autogen), or the flattened hedge rows at the edge of the field, and it just kills it for me:

Another example from later in the clip, with sharply detailed autogen sitting on blurry terrain with squashed cars, squashed buildings at the lower right:

I know I'm cherry-picking here, but I can't help noticing these things. And sure, you wouldn't notice this at high altitude, but the times when I'm paying the most attention to scenery is during takeoff and landing, where I do notice these things. Or if I'm flying a helicopter and there isn't a need to climb very high on a flight. 

This isn't a problem that orthos will ever solve, because even if you have insanely high resolution, flying at any major angle to an object on the ground will reveal that it's flattened. Marinas are a special eyesore, because nobody ever goes in and covers up all the flattened docks and boats with hand-placed objects. 

This doesn't even get into the issue of different sunlight angles competing with baked-in shadows. If you only fly at noon on a sunny day with no clouds, I guess it doesn't matter, but that's not how I fly. So I'll keep flying over "plausible" landclass + autogen that responds to changing conditions of weather and sunlight as a personal preference.

Looking towards the future, I have a feeling that whatever Austin has planned for procedural season changes will also be a better fit for their default terrain + OSM, than it will be for orthophotos.

So how does it work? Does it literally somehow "read" the ortho and put trees based on coloring or something?

1 hour ago, bonchie said:

So how does it work? Does it literally somehow "read" the ortho and put trees based on coloring or something?

Basically, yes. It will scan all the DDS files in your Ortho4XP ortho directory and generate trees based on color with some simplification and smoothing. It's pretty simple and nothing that hasn't been done before (There is a similar tool for FSX)

1 hour ago, alpilotx said:

By the way ... for the USA, I think you might - instead of doing it algorithmically yourself (with all the pitfalls included) - really think about falling back on NLCD data for example, which is really detailed and should be aligned quite well with most of the orthoimagery (is also very high res)

I've actually started using this approach using the NLCD raster data as it's quite detailed and has simple forest/vegetation classifications. In addition, I've found a few states which provide some even higher resolution landuse data which can be used (I've already tried the California vegetation one, and it's pretty good). 

 

 

 

1 hour ago, Paraffin said:

I just can't get past the artifacts at low flight levels -- the glaring contrast between razor-sharp detailed 3D autogen buildings and the blurry ground they sit on.

This is a mistake I think many airport sceneries have when using orthos for their ground textures. Many developers don't edit out the buildings or cars and it really ruins it. This is why I really like the work of BetiX and some ORBX airport devs because they make their own ground textures or carefully add AO and edit the orthos so they work naturally with the buildings. However, for a large scale area, this would be a huge amount of work.

2 hours ago, tonywob said:

Here are a couple of shots of a few tiles around Portland and Oceano using the autotrees

Looks quite good Tony, looking forward to trying it.

27 minutes ago, tonywob said:

I've actually started using this approach using the NLCD raster data as it's quite detailed and has simple forest/vegetation classifications. In addition, I've found a few states which provide some even higher resolution landuse data which can be used (I've already tried the California vegetation one, and it's pretty good).

Any plans to share this as well sometime?

I'll upload the California one to x-plane.org when I can. 

 

8 minutes ago, tonywob said:

I'll upload the California one to x-plane.org when I can. 

Awesome, thanks!

Thanks Tony, looking forward!

Bob

Really looking great Tony!

For me it isn't too much about making sure the trees are in the right spot, it's just about making sure the trees aren't in the wrong spot. 

 

Looks like your efforts here could be a very nice addition!

23 hours ago, tonywob said:

 

Here are a couple of shots of a few tiles around Portland and Oceano using the autotrees

 

IMO those look fantastic Tony.  Are you planning on making a tool available to do this (like w2xp) or are you just planning to upload different regions to the .org?

Tried it again in a different area and I'm warming up to them. The overlays just need more variety. Hopefully as LR expands the database for autogen objects, that'll improve.

Gonna add W2XP on top now even though US coverage isn't very good. At least it'll add some landmarks. Do you recommend using the VFR world landmarks package as well with orthos?

20 minutes ago, bonchie said:

Tried it again in a different area and I'm warming up to them. The overlays just need more variety. Hopefully as LR expands the database for autogen objects, that'll improve.

What I'd like to do next is add W2XP overlays on top as well. I'm a little confused on exactly what I need. I'm wanting it for the US, so download W2XP America obviously, but should I also add VFR Landmarks? And what order would they go in in the scenery library?

If you are doing the western US, I would recommend using US West "Real Life" as it adds "locally flavored" objects.

If you do use that, the scenery order would look like:

-Airports, libraries, etc

-US West

 -W2XP America

-Ortho4XP overlays

-Ortho4XP tiles

-Other mesh

If you are talking about this VFR Landmark scenery from Simheaven, I don't believe you need it as it should be included in W2XP America. 

 

Tried to get W2XP working but I'm getting missing object errors that crash the sim.

I googled and most said to download the latest world models package. Is the one on simheaven not the latest? I'll search around.

Edit: Fixed issue. Was missing an unspecified library (FF_library)