Ground Projection: Corona vs. V-Ray

[kosso_olli]

Post 2/2

Now, lets alter the perspective, to see the rest of the scene. As you can see, there is a severe distortion.


Anything getting close to the horizon line is distorted because of the difference in size below and above the horizon. The lower half below the horizon is correctly scaled, the upper half above the horizon is incorrectly scaled (too big).



To eliminate that, users need a parameter to scale the projection above the horizion line. The same procedure in Corona is working fine, because you have that parameter.

Additionally, a parameter to control the distance where the blend is taking place would be very helpfull. Right now, the "blend edge" is too close to the camera. For some environments, it would be nice to be able to shift the blending edge further into the distance.


I hope this illustrates the problem I am having. A solution to this should work for Jonas as well, as it is basically the same issue.

[^Lele^]

Can i see your camera's position, and the map settings for the ground projection?

[kosso_olli]

The ground projection settings can be find in post 1/1, but here they are again, along with camera position.


And no, setting the ground projection to the camera height in scene does not help at all. First, it does not solve the distortion. Second, the scaling on the lower half gets messed up. Been there, tried all of that years ago.



Your way of thinking is flawed, the camera position in scene does not correspond the the camera height of the ground projection.

EDIT: Inserting the values for X and Y as well, the distortion goes away, but scaling is still wrong. The size is now too big, above as well as below the horizon:

[^Lele^]

That is wrong.
Set the settings as i showed you here in the second image.
In your money: in the map settings, set x and y to the camera's x and y. Height, to your camera's Z.
The height has *nothing* to do with the tripod. That's zeroed by the spherical projection.
It's rather the height of your camera from the (projected) ground plane.

So, if you don't need an arbitrary scaling, and you don't need VRed's options, we're good?
EDIT: it occurs to me i could script you a way to connect the map's parameters to the camera's. But you can do that with the max UI, right?

[kosso_olli]

That is wrong.
Set the settings as i showed you here in the second image.
In your money: in the map settings, set x and y to the camera's x and y. Height, to your camera's Z.

See, I did that in the third image in post #63. It does not work. Also, the result gets worse the higher the camera is positioned in scene.

The height has *nothing* to do with the tripod. That's zeroed by the spherical projection.

It does. It determines the correct scale of the projection. Any deviation away from the tripod height results in a wrong scaling for the ground plane projection.

So, if you don't need an arbitrary scaling, and you don't need VRed's options, we're good?

I do need the scaling! We're not done here yet!

[kosso_olli]

Further images to show the flaws of your solution.

Here with the settings I believe to be correct. Size and projection look okay to me.



With the camera translation values typed into the corresponding fields of the VrayHDRI. That just looks wrong.

[^Lele^]

See, I did that in the third image in post #63. It does not work. Also, the result gets worse the higher the camera is positioned in scene.

It does work here, *invariably*, in the way i explained.
Anything other than 0.0 in the map's Z and the camera Z in the map's "radius", and the projection (duly) distorts.

I see your point, perhaps then a change of thought is needed.
EDIT: nope, wrong.

[kosso_olli]

It does work here, *invariably*, in the way i explained.

Please provide a simple scene file, because there is no way to comprehend where the error is.

The fun fact is that Corona does *not* change the projection scale changing the radius.
That only changes the size of the sphere, and as such the lighting (and scene bBox?).

I'm totally fine with that. That is exactly what is needed.
I don't care what is behind all this, if it is a sphere, a dome, an infinite space. Heck, I don't even know how it is in V-Ray. All I can say is that the results right now look nowhere near as good as Coronas.

Unit setup is correct on our end, by the way. We're working in Centimeters, and the values of the reference plane etc. match accordingly. If there is any internal conversion error going on in V-Ray (Imperial to metrics, or Centimeter to random body part length), that would be out of my scope.
The slight warping of the lines on the ground is to be expected for the spherical projection. Hence the proposed "distance" from scene origin to tell V-Ray where to blend between groundplane and background.

[Pixelcon]

You can find a ton of tutorials on the internet.
Or use Corona.

Remarkable how the tutorial title begins with 'Solved'...

[kosso_olli]

It does work here, *invariably*, in the way i explained.

It does not work at all. By typing in the camera translation value, you basically negate the ground projection. Take a look at the next three images.

With ground projection and radius=tripod height. Distorted, but the car is standing on the ground, scaling of the ground is correct.




With the camera translation values typed into the ground projection. Scaling is off, car appears to be floating in space.




Ground projection turned off entirely. It is exactly the resulst as per your workflow. So this is plainly wrong.



How many more images do you need before you admit that this can NOT be a solution to our problem!

[^Lele^]

Wait up, i tried a fifth map, and that was fubarred.
Then i tried a sixth, and it was fubarred too.
So, this is how i understand it, specifically for v-ray.

The x-y-z are the camera's, the height is the tripod's.
So, with everything centered in the origin (camera 50 units above ground), the map "position" is set to [0,0,50] (that is the projection origin.).
The height of the tripod is whatever was set into the captured HDRI.

*then* i am as distortion free in V-Ray as i am in Corona.
The projections do not match exactly, but both are usable.
Problem on our side is the tripod height is anyone's guess (especially on random internet images), while Corona admittedly works slightly better with its defaults (i.e. position set to 0. *some* camera height at around 80cm).
There is still a ton of guessing for camera height and sphere radius to go on (so to ensure one isn't outside of the sphere, via setting camera height), to get a distortion-free result, with either solution.

I hope we'll manage to go over this and sort it out properly (with a projection to arbitrary geo approach, like iRay/VRed), so to make both that wee bit better than they are.

[JonasNöll]

kosso_olli : Sent you a private msg with the files from the demonstration images!

[^Lele^]

How many more images do you need before you admit that this can NOT be a solution to our problem!

As many as it takes.
In Corona, you can switch to a perspective, and zoom out from the "negated" ground projection.
The second you move the camera you'll notice the projection was in full swing, what was negated was the distortion (which you call "scale".).
You can't do that in V-Ray, as by now you'll know, to not break the lighting the projection is to a sphere with infinite radius.

Again, the settings are subtly different between the two, but either is capable of essentially the same type of job.

[JonasNöll]

Lele, the offer stands, I send you the demonstration files and you can show that with your approach you can reach the same or similar result like in Corona that looks obviously less distorted. I mean it would be fine to just say: "Yeah sorry guys we can't do that in V-Ray because of X,Y,Z and also dont plan to implement any feature because because of X,Y,Z"
Correct me if I'm wrong but you are saying: That it CAN be done in V-Ray, probably through more complicated ways but it can be done. So if you could just offer this solution to me that would be really helpful, because so far I just got 5 pages of lecturing and no real solution . Instead of all this time spend replying why not just provide the solution in 5 minutes and everyone would be satisfied. I also sent the file to Chaosgroup support and they looked at it and couldn't offer any solution.

[^Lele^]

Do send away, then.