Ground Projection: Corona vs. V-Ray

[fixeighted]

Lele is Lele - makes for an intensely interesting thread
It's a very good argument, on both sides, and has led me, personally, into testing and other areas that I would not have thought about without such a vigorous discussion.
Freedom of speech is cool and honestly, I am not offended by anything in this thread...it's just polarised thinking it seems....the ideal is surely inbetween...
I reckon having Lele in the mix, rather than any sanitised customer facing platitudes, makes for a better mix.

[^Lele^]

Lele, for someone employed at chaos group you are always acting at odds with what was famously great support. The way you respond to people is gross. Have you seriously never been pulled aside and told to dial it back a bit?

Besides from you, not once, so far.
I very often ask for vetting of my posts, so often i post with permission, to the word.

Regardless, curious as to where the reply is "gross".
Also, should you feel the issue isn't addressed, i'd like to hear why: center the projection on the camera, and adjust height/Z, depending on the effect one wants to achieve (moving projection or texture center to flatten the ground), seems proper.
I'm all ears for another coherent solution.

[fixeighted]

Anyway, to go back to Jonas's initial question; with every hdr I tried I can't see any distortion at all if set up as suggested.

Naturally there are issues with reflection but I overcame these (depending on specific hdr) by using mapped geometry for the buildings and floor as well as any other prominent objects that need to be reflected when close up.
The attached took me an hour or so and shows those methods. Obviously quick/dirty but makes a case for ease. A little extension onto other parts of the image allow for animation away from 0,0,0 also.
Thanks to Dabarti for practically the only vid out there of a great technique for mapping the geometry https://www.youtube.com/watch?v=ZEpc...hannel=Dabarti

Maybe I misunderstand what others want to do but if this is it then it's not a one-click solution but isn't too hard either

[Neilg]

Besides from you, not once, so far.
I very often ask for vetting of my posts, so often i post with permission, to the word.

Fair enough
it wasn't an issue with the method you were proposing, I feel like many of your posts 'helping' people are argumentative and confrontational when they don't really need to be.

I'll bow out, looks like this is a me issue, let this get back on topic.

[^Lele^]

Fair enough
it wasn't an issue with the method you were proposing, I feel like many of your posts 'helping' people are argumentative and confrontational when they don't really need to be.

I'll bow out, looks like this is a me issue, let this get back on topic.

I am often dry and direct, and my role is to be argumentative (in the strict sense of the word, not in the negative connotation.).
As i explained, it's to ensure we get to an *actionable* request, lest it'll be lost into the to-do list, or the bug reports, or worse still consume us precious human resources for no appreciable result.

That i may be often wrong, either in tone, or content, is unquestionable.
For either i'll take founded criticism and adjust to do better the next time.
I'll reread the thread and try and understand where my tone wasn't proper.

But please do not read my role as antagonising of or flippant towards user needs, for it's the very opposite.
I dig, so others can build.

p.s.: the docs are scant on the ground projection, which gave rise to all of this debate. We will need to do a bit better there.

[JonasNöll]

Also, should you feel the issue isn't addressed, i'd like to hear why: center the projection on the camera, and adjust height/Z, depending on the effect one wants to achieve (moving projection or texture center to flatten the ground), seems proper.
I'm all ears for another coherent solution.

Hi, so apart from the tone where I can personally agree to what Neilg was mentioning, from my account this original issue is not really solved as the solution proposed required to offset the center of the whole HDRI and only works from one specific angle. Not sure if you would wire the camera position into the coordinated of the HDRI you could probably make it possible that you could move the camera around with this approach, but the center of the HDRI would always be moving depending on where the camera is.
So from my side this doesn't really be a solution compared to the approach in Corona where my model would always stay the at the center of the HDRI and I can freely move a camera around.

[kosso_olli]

Anyway, to go back to Jonas's initial question; with every hdr I tried I can't see any distortion at all if set up as suggested.

I tried to make my point a few times now in this thread, but somehow I fail at explaining what result I would like to get.

The biggest problem with the workflow Lele suggested is scaling. By moving the origin of the HDRI sphere to the camera origin, you basically ignore all the scaling, meaning the projection on the floor is the wrong scale, as well as anything above the camera horizon.

Look, the workflow we employ might be highly specific and somewhat niche in V-Ray, but it's like this

- we get a highres backplate with a stand-in car, along with detailed reference for the shot. This means data for camera model and sensor, distance to the object of interest, camera height on tripod, measurements of scene-specific items (curbs, parking lines, height of building windows/doors etc.)
- based on this data, the camera is accurately matched, so that the car can be replaced with the rendered one.
- along with the measurements we either get an HDRI by the photographer, or we shoot it ourselves while on set. The HDRI is shot at exactly the same location where the car was standing. Depending on the lens used to take the backplate, the car can be standing up to 50m away from camera with a standard lens, and further away for shots captured with a camera on a drone
- it is done this way to make sure that we get accurate reflections in the car and to get the right size on the ground after using the ground projection feature. It also means that we do not have to remodel any part of the scene (in 99% of cases. There are situations where this can not be avoided). Having to remodel the environment might take 1-3h depending on complexity, for some shots it is impossible altogether (rig-shots, shots with lots of vegetations, i.e. bushes or trees)


With Lele's approach the spheres origin (i.e. the tripod location at acquisition) is shifted to the in-scene camera location. While this does not mean that the reflections are wrong in all cases, it means that you are loosing quality in the reflections, because the area around the teapot gets stretched too much:




But what is more important is the scale of it all. It gets unusable for drone shots like this. I used the same settings as recommended by Lele. But as you can see, the ground projected using that approach is totally out of scale. It is not when using the tripod height for camera radius.

Lele's way. Scale is out of order, by a big margin



Our approach. Scale is okay, albeit the slight pinching towards the edges.



For all cases teapot means car. Also, using the tripod height as radius and values for XYZ=0 means we can render the same car from multiple camera angles without touching the projection settings again. Everything stays locked in place in relation to the teapot/car.
It is not when using the camera-translation approach.

I hope this makes it a little clearer. It is not so much the distortion that makes trouble, it is the lack of an option to control the global scale. As soon as we have that option, the distortion can be minimised to a great extend because you can balance ground projection with scale above camera horizon.

And one last thing: I have to admit I lost temper last week in some of my posts. I already apologized to Lele, and I do the same for anyone feeling offended: Sorry!

[kosso_olli]

Not sure if you would wire the camera position into the coordinated of the HDRI you could probably make it possible that you could move the camera around with this approach, but the center of the HDRI would always be moving depending on where the camera is.

Exactly, your model would appear to be floating in space, even when wiring the camera position to the ground projection XYZ.
Like there was no ground projection at all.

[^Lele^]

You *need* to lift the projection from the neutralised version.
Either with tripod height (to shift texture) or with projection Z (to shift projection).
Then you get your floor squished as much as you want it, without side-to-side skew.
Scaling will change accordingly, your lines on the ground parallel, instead of warped.

[kosso_olli]

Either with tripod height (to shift texture) or with projection Z (to shift projection).

We do not need to shift projection with these values, ideally they are all set to 0.
This only needs adjustment when the HDRI was not take exactly at the same place where the car was. It is best to avoid shifting the projection, more than 50cm in either direction or you loose quality because there is even more stretching.

[^Lele^]

Do you actually Light with the ground-projected environment?

[kosso_olli]

Do you actually Light with the ground-projected environment?

Yes, we do. It does make quite a big difference if the car is standing right next to a bright spot on the ground. You will get the correct bounced light in the car, then. It is most visible in shots where one half of the car is in the dark, while the other half is hit by sunlight.
Here are two examples. These are only lit by HDRI, not additional light sources used.

Here you can see the transition from the areas in the shadow (the front fascia) to areas with bounced light from the ground (sills and rocker panels etc.). The corresponding HDRI was shot excatly at the shadow/light transition on that street.



This one is nearly the same situation, only HDRI which already had the shadow/light areas in it. As a result, the front bumper gets bounced light on the right side, while the left side stays dark.

[^Lele^]

I asked to understand if losing MIS was important, i gather it is?

[kosso_olli]

Oh, okay. MIS is important, yes.
I remember getting excessive noise and fireflies in cases where we had to quickly remodel some parts of the environment and used that for reflections with a LightMtl and the HDRI as texture applied to it. It was in 3.6 I think, so those issues might be fixed already.

[^Lele^]

Eh, I hear you.
We'll see what is possible within these constraints.