Trees, trees, trees, trees, trees, trees,...

[Dominique_K 10,133]

... lots of trees, 1,5 trillion  of them ! 

But which trees ? How many varieties ? This is something that is not clear to me. We saw acacias with the giraffe herd but the information is scarce about the vegetation differentiation. This is a question  that some future previews would be welcome to address.

[MatthewS 761]

Yes FS2020 needs variety to look "real".

Don't spoil the natural variety of the satellite imagery by populating it with too few tree models.

[bonchie 2,305]

I'm not exactly sure what they are doing, but I think having an FSX mindset of needing "variety" via libraries may be a mistake.  

For all we know, trees may be produced the same way buildings are being done, i.e. via AI scanning orthos and coming up regional varieties dynamically based on real life locations. 

Edited October 11, 2019 by bonchie

[Dominique_K 10,133]

1 minute ago, bonchie said:

I'm not exactly sure what they are doing, but I think having an FSX mindset of needing "variety" via libraries may be a mistake.  

For all we know, trees may be produced the same way buildings are being done, i.e. via AI scanning orthos and coming up regional varieties dynamically based on real life locations. 

I am not sure that I understand what you mean by "mindset of needing "variety" via libraries".

To feed its algorithms, the AI would  need a database of correspondences between locations and probable types of trees and another one of shapes and textures to construct the trees.

As for many things, the FS family (FSX+P3D) doesn't do a bad job (i've planted so many trees myself with IS !) but it could be much better. 

[skully 1,045]

When I read that Azure had tagged 1.5 trillion trees on our planet in two days- I had to laugh out loud at the sheer awesomeness of such a feat.

[bonchie 2,305]

40 minutes ago, domkle said:

I am not sure that I understand what you mean by "mindset of needing "variety" via libraries".

To feed its algorithms, the AI would  need a database of correspondences between locations and probable types of trees and another one of shapes and textures to construct the trees.

As for many things, the FS family (FSX+P3D) doesn't do a bad job (i've planted so many trees myself with IS !) but it could be much better. 

The FSX mindset would be hand drawing regional boundaries and then hand assigning tree types via a library to those regions. 

I'm saying that if the AI is doing it automatically, and it seems it is, there's zero reason to believe it won't be far more accurate than what we've had before in choosing the right types of trees to place in the right spots. We already know the AI was used to find the original locations of all 1.5T trees. 

[Dominique_K 10,133]

47 minutes ago, skully said:

When I read that Azure had tagged 1.5 trillion trees on our planet in two days- I had to laugh out loud at the sheer awesomeness of such a feat.

Suppose they are wrong of about 5%, just 5%, which is not much. 75 billions of trees missing or too many 😃 

 

13 minutes ago, bonchie said:

The FSX mindset would be hand drawing regional boundaries and then hand assigning tree types via a library to those regions. 

I'm saying that if the AI is doing it automatically, and it seems it is, there's zero reason to believe it won't be far more accurate than what we've had before in choosing the right types of trees to place in the right spots. We already know the AI was used to find the original locations of all 1.5T trees. 

OK. The present system is indeed a little coarse in this regard..   

The chance  that the AI puts the  model of the right tree in the right place depends however  on how rich its databases are. So far I have not seen much (except the acacias in Africa.). I look forward to get more infos in due time.

Edited October 11, 2019 by domkle

[bonchie 2,305]

1 minute ago, domkle said:

Suppose they are wrong of about 5%, just 5%, which is not much. 75 billions of trees missing or too many 😃 

 

OK. The present system is indeed a little coarse in this regard..   

The chances that the AI puts the  model of the right tree in the right place depends however  on how rich its databases are. So far I have not seen much (except the acacias in Africa.). I look forward to get more infos in due time.

Except there are no starting databases. That's the old mindset I'm talking about. They aren't using a forestry database like someone tried to do in X-Plane. The AI is creating its own database from scratch. Unless ortho for a region simply doesn't exist to read, there's no reason to think it'll have any prejudice, whether it's Africa or California. Any region with ortho should receive the same amount of attention because the AI doesn't know any better. 

That's my take given how they've explained the AI is reading tree locations. I could be wrong. 

 

[MatthewS 761]

6 hours ago, bonchie said:

For all we know, trees may be produced the same way buildings are being done, i.e. via AI scanning orthos and coming up regional varieties dynamically based on real life locations.

Yes, maybe when they analyse the satellite imagery they also incorporate the subtle variety in color, some trees lighter or darker foilage than others, maybe a whole group of trees suffering some "dieback" etc etc. 

If the natural variation in color apparent in the imagery is translated into the sim then that will help make it look real.

It already looks great anyway, but hopefully they never give up tweaking the fs2020 world.

[aleex 551]

Honestly, Idk how azure is making those trees, but I prefer the photogrametry ones a million times over those "Azure all the same" ones. At least the first ones has different forms and heights, and match the real shape. And for me, those are more leafy.

I mean, look at the difference... It's noticeable. MS, you already had the work done, why make it worse?

Edited October 12, 2019 by aleex

[aleex 551]

Another example

 

 

Don't get me wrong, the sim is still looking excellent, but in my opinion the current autogen trees are worse than the photogrametry ones. I know the limitations regarding photogrametry at close distance or low parts of the trees... but nothing is perfect.

[Dominique_K 10,133]

6 minutes ago, aleex said:

 

but nothing is perfect.

 No, I agree, but trees are important, shape and variety, for the simmer who flies under 10k feet and from/to small airfields. They'd better do that right. 

[sightseer 895]

you can't really make comparisons using those photos due to different perspectives and very different lighting conditions.

[aleex 551]

1 minute ago, sightseer said:

you can't really make comparisons using those photos due to different perspectives and very different lighting conditions.

I'm not comparing the scenes, I'm comparing the trees, which are absolutely different.

But you can always use the earth view app to compare by yourself... It's also a funny app to use for take a look at the included 3D cities.

[Nedo68 533]

Photogrammetry trees looking only good from distance, close up they look ugly like inflated balloons.

You need to place better 3D Tree modells manually or with Azure AI.