Ground Projection: Corona vs. V-Ray

[fixeighted]

Aah...that is a very cool technique!...I hadn't thought of that.
If I have it right of course...I mapped the hdri map to the floor, inside a light material and then made the floor a matte object.
Improvement all round. Please tell me if I missed anything.

Oliver, a question about that first image above; did you use this method in that?
Also, I'm assuming you had shadow casters across the car and simply matched the existing road shadows?
It's how I would approach it but always nice to hear alternatives. I'll understand if you don't want to reveal tricks

What I don't understand is about MIS and why that is an issue, so clarification would be great please.
I left it on, as without it it creates fireflies; so in which situation would it be useful to untick it?

On the original discussion; I have played with this in Corona and Vray for the length of the post and I have to argue that
having a similar option in Vray as what Corona offers would be great. Neither solves certain issues with reflections but
Corona's is definitely a great rapid fix in a lot of cases.
I would hope that some sort of future AI involvement can further expand the possibilities

[^Lele^]

Aah...that is a very cool technique!...I hadn't thought of that.
If I have it right of course...I mapped the hdri map to the floor, inside a light material and then made the floor a matte object.
Improvement all round. Please tell me if I missed anything.

That *is* the way projections are done, and have always been done.
Matte Painters are rightfully paid their weight in gold.

MIS has to do with sampling: if it's absent, sampling will be poorer.

As for Corona, what exactly is the bit working better (i am truly trying to find out. my tests show no appreciable difference so i am missing something obvious.)?

[kosso_olli]

Aah...that is a very cool technique!...I hadn't thought of that.
If I have it right of course...I mapped the hdri map to the floor, inside a light material and then made the floor a matte object.
Improvement all round. Please tell me if I missed anything.

You can do it like this, yes. But at the cost of noise. Especially for night shots, where there are loads of tiny little light sources in the HDRI, you might get noise and fireflies all over in the reflections. You could clamp the HDRI for the reflection to get rid of the noise, but that also kills the information for Glare and Lens Effects.
Also, you have to find a way to blend the ground and the LightMat with the rest of the environment. Gradient in opacity etc. It works in a lot of cases, but it needs quite some fiddling.

Oliver, a question about that first image above; did you use this method in that?
Also, I'm assuming you had shadow casters across the car and simply matched the existing road shadows?
It's how I would approach it but always nice to hear alternatives. I'll understand if you don't want to reveal tricks

Just the spherical HDRI was used, with ground projection in the VrayHDRI. In this case it worked. No bounce-cards or additional light sources.
There are two shadow casters in there, yes: One for the lantern, and a box. But still, the front would stay dark if these objects were not there, because the shadow area was already captured in the HDRI, thus darkening the front because there is no bounced light.

[fixeighted]

No, not that sort of matte
I meant about using the light material to light the floor (which is a plane, not the dome) and so get the bounced light from the correct bits of the HDRi
I wouldn't have thought about it if you had not asked him whether he was using the ground projection to light.
This post has been very useful so far, especially as I need to present the Range Rover I built recently.

MIS; so why is there a tick box....what situation would require it be unticked?

It's not that Corona works better, just differently...so in Oliver's case, he says the reflections are bad because they come from an infinite sphere and so are naturally a bad scale, therefore having a smaller sphere would make
them perceivably closer, smaller, more acceptable. Which is what the dome option in Corona allows a bit nicer control of, though by any means not perfect.

Just a thought - for close reflections, wouldn't it be easier to take large frame flatter panoramas on set; maybe drone captures of the ground also, rather than rely on just one main hdr?

As an aside, and in trying to fully understand what were the issues around this whole topic, I tried to imagine what would be the best and quickest/cheapest method of
capturing hdris that'd be better usable in these situations.
Then I happened to remember Mari and though that might help; but it's years since I even looked at Mari and I couldn't quite grasp the essence of what I was thinking in order to describe it here.
Then I stumbled upon this just now https://www.fxguide.com/fxguidetv/fx...-mari-and-hdr/ and it is what I was thinking, more or less.

This was in 2013 so it's probably a lot cheaper/easier to do this for shots where one hdr isn't going to be enough.
It doesn't solve this particular thread's problem but it is maybe a good solution for those who need to do these shots on a regular basis, which I do not

[fixeighted]

Thanks for clarifying that Oliver.

With my sphere test I didn't have any issue with blend, as in these images.
The light material is defaults with the same hdr.
I'll try it with some other hdr shots - maybe it was chance...
How did you approach it if different?

[^Lele^]

No, not that sort of matte
I meant about using the light material to light the floor (which is a plane, not the dome) and so get the bounced light from the correct bits of the HDRi
I wouldn't have thought about it if you had not asked him whether he was using the ground projection to light.
This post has been very useful so far, especially as I need to present the Range Rover I built recently.

Ahah, it's one and the same "Matte Painting".
Reprojections (check out the max Projection modifier, and camera map per pixel map, and the Camera Map Gemini plugin) are the order of the day in nigh any 3d endeavour where (animated) integration is the goal.
I'm *super* glad this whole thing has turned out to be food for thought.

MIS; so why is there a tick box....what situation would require it be unticked?

That's a relic, It shouldn't ever be turned off.
In general Multiple Importance Sampling is a boon to rendering.
A dome allows for (much) higher optimisation than a light material.

It's not that Corona works better, just differently...so in Oliver's case, he says the reflections are bad because they come from an infinite sphere and so are naturally a bad scale, therefore having a smaller sphere would make
them perceivably closer, smaller, more acceptable. Which is what the dome option in Corona allows a bit nicer control of, though by any means not perfect.

I haven't found that the dome size changes anything at all in the size of the projection on screen.
A fixed-size dome will change lighting (your sun will not be a parallel source, f.e.), and it will change how glossy a reflection (no. all reflections from the same dome! Forget parallax or occlusion.) will look, but *not* the viewed size.
EDIT: While it's true the reflections change apparent size, that only happens around extremely small dome sizes. Sizes which would make, f.e. outdoor lighting with a sun unfeasible (i.e. if one managed to squeeze the scene in it, it'd still look like a spotlight, rather than a parallel light.). Any size above 1000 units will have diminishing returns. Further to this, stretching due to the ground projection is always very visible in the reflections.

Just a thought - for close reflections, wouldn't it be easier to take large frame flatter panoramas on set; maybe drone captures of the ground also, rather than rely on just one main hdr?

To have proper integration, as you point out, one needs to map and move actual geo inside the scene.
Which brings me to my undying qualms: it's the *first* time i hear someone wanting to pull together so many different aspects of scene construction into one simplistic solution.
Me and the teams of people i worked with, be it in film, or cars, or archviz, went to extreme lengths to cover our bases with multiple HDRIs, a ton of backplates, and tenfold those in pictures with which to map geometry, *specifically* to have control over what does what in a scene.
Because a single dome had never remotely cut it, much less so when used for lighting, direct viewing, and reflection/refraction.
That ground projection should be the panacea to all ailments sounds to me way too optimistic.
Or maybe we were all a bunch of nerds, overcomplicating what was obviously very easy, and we all obviously missed out on.

As an aside, and in trying to fully understand what were the issues around this whole topic, I tried to imagine what would be the best and quickest/cheapest method of
capturing hdris that'd be better usable in these situations.
Then I happened to remember Mari and though that might help; but it's years since I even looked at Mari and I couldn't quite grasp the essence of what I was thinking in order to describe it here.
Then I stumbled upon this just now https://www.fxguide.com/fxguidetv/fx...-mari-and-hdr/ and it is what I was thinking, more or less.

Yeps, precisely my point.
At around minute 11 he makes the remark, as he does in the comments, about the *extra* work to do besides the HDRI.

This was in 2013 so it's probably a lot cheaper/easier to do this for shots where one hdr isn't going to be enough.
It doesn't solve this particular thread's problem but it is maybe a good solution for those who need to do these shots on a regular basis, which I do not

Nothing of substance changed since then (fair warning, that's also when i stopped doing production. So what do i know.).
It's a labour of exceptional preparation and skill (which is why i have always lived in utter awe of Matte Painters and their insanely broad skillset.), one i fear no single-click solution will ever supplant.
And my whole point about this thread was precisely that: distortion like the one shown time and again in here is *ever* present in ground reprojections, and it will look worse the farther a camera center is from the projection one.
Picking specific instances where it's slightly less in one application and slightly more in the other isn't actionable, as both suffer the same issue, and there is no "copying" to be done to solve anything at all.

Alas, given MIS is needed, we won't be able to do what VRED does on arbitrary geo (which, to be fair, can already be done with the standard Max tools. If with some precision issue here and there.).
We may be able to blend better the joint between the dome and the plane, but even then, it will not allow people to achieve what it is they seem to be wanting.
Or at least this is how i see it.
vlado will surely know better.

[kosso_olli]

We may be able to blend better the joint between the dome and the plane, but even then, it will not allow people to achieve what it is they seem to be wanting.

For that reason, I suggested the introduction of a size value or multiplier for the size of the upper half of the dome.
My biggest problem is not so much the distortion, but it is the appearance of the upper part of the dome, because it is infinitely far away, and therefore too big.
All the rest can be done with the ground projection as is.
The dome will never be directly viewed. If it was, we would remodel the environment, and we have done this in the past. But for lighting and reflection, that "multiplier" would be a quick and easy solution.

Basically, the thread from three years ago sums it all up: https://forums.chaosgroup.com/forum/...rrectly-scaled

[^Lele^]

The dome will never be directly viewed.

So what i said about centering on the camera is relevant insofar as you center it on the geo.
Just do not look at the ground projection from a different point of view but that of the reflections.
If you do, as you have shown to us in the samples, the ground will "shrink" and be pinched, as it happens to ground projection when the camera is back from the center (conversely, it grows if the camera is pushed in).
The reflections should be fine, provided the probe is centered.

In this case, yes, a fixed-size (hemi)sphere will make the size of the projection matter, in relation to the size of the reflections, but the issue with the distortion of the *directly* viewed probe remains.

[fixeighted]

tldr Nothing is perfect and best idea is to do custom projections where needed. Certainly for animation this is unavoidable for the most part.
Still would vote for inclusion of a Corona-like implementation for the following reason....not to annoy you I promise Just an example...

This image was done in seconds with Corona set to dome...very small radius.
It's only the perspective view and I can freely move all around/above it as long as I stay in the dome.
With no distortion other than what the size of the dome creates naturally (the same relatively as does Vray, just scaled)

It indicates that the reflection of the firehose in the roof and the windows are where they should be and look around the right size.
The same ease of setup and immediate use isn't possible with Vray at the moment.
With Vray we need the ground projection disabled to get a proper background, yet that can't be reflected properly as it's too far away. And it looks too big
Using ground projection means we need to use a camera that is locked to xy 0 and a fixed height
And unfortunately the reflections still look wrong, even compared to my quick and still incorrect version that took seconds
Was going to post the vray vesrion but just crashed.

It just a nice quick and instantly useable feature I reckon.

[^Lele^]

I get distortion if i move, and the vertical lines are very warped even in your picture, but i do see your point, sure.
In fact, a fixed size dome is what is used to make reflection probes, so it's not news in this respect.

Corona's implementation has two issues, however:
One is with how it's enabled, currently, and the fact it can turn a scene into being unusable (lighting off, or changed completely) with a single click in a *texture map*.
The other is with lighting (only reflection probes had fixed size, eh.), as the small size will make your sun a broad spotlight.

For this, i'd *much* prefer we made a dedicated "environment" system (similar to the iRay one, or better still VRED), so that what was doing what was a bit clearer.
From a single window/tool, i dream of being able to cater for most environment-related needs.
You want a squashed sphere as reflection dome with ground projection? fine.
You'd prefer a repro of the HDRI dome on some custom mesh' UVs? Pick the map, pick the mesh, project.
Need to light? Here's a dome to do IBL with.
Want a backdrop to your render? Tick this and load the map.

I'm not going to be the one coding it, though.

[kosso_olli]

This image was done in seconds with Corona set to dome...very small radius.
It's only the perspective view and I can freely move all around/above it as long as I stay in the dome.
With no distortion other than what the size of the dome creates naturally (the same relatively as does Vray, just scaled)

It indicates that the reflection of the firehose in the roof and the windows are where they should be and look around the right size.
The same ease of setup and immediate use isn't possible with Vray at the moment.

Thank you. This is a perfect demonstration of the difference. Something like this is needed in V-Ray, too. We'd replace the directly viewed background by a camera backplate, but the reflections would look so much better and more convincing. Not like a toy car in a real-world scene.

Corona's implementation has two issues, however:
One is with how it's enabled, currently, and the fact it can turn a scene into being unusable (lighting off, or changed completely) with a single click in a *texture map*.
The other is with lighting (only reflection probes had fixed size, eh.), as the small size will make your sun a broad spotlight.

Well, if the scene is created exactly with that intent, I see no problem for the different lighting. Maybe print a warning to the checkbox, similar to the additive mode in BlendMtl.
Also, in the Corona image by fixeighted, the sun is still creating sharp and strong shadows. The size does not to have an effect on lighting that much.

For this, i'd *much* prefer we made a dedicated "environment" system (similar to the iRay one, or better still VRED), so that what was doing what was a bit clearer.
From a single window/tool, i dream of being able to cater for most environment-related needs.
You want a squashed sphere as reflection dome with ground projection? fine.
You'd prefer a repro of the HDRI dome on some custom mesh' UVs? Pick the map, pick the mesh, project.
Need to light? Here's a dome to do IBL with.
Want a backdrop to your render? Tick this and load the map.

As long as there is anything to control the size for the upper hemisphere of that squashed dome, I'd be happy...

[fixeighted]

Until you typed that last line I was just beginning to think of the vast amount of gifts that everyone would bestow upon you.
Oh well, you could have been the proud recipient of a pair of these natty helpers!

Anyway, for sure it won't stand up to some scenarios but will be a real help in many.
I see what you mean about sun size and that would be a situation where it wouldn't help.
However, many scenarios are indoor ones and so that wouldn't matter so much. Some outdoor ones would also be fine if the situation was right.
I think the best reason for having it is that it's just another tool, which clearly has its uses.
If it was my renderer (oh to be bright enough to write one) and people were choosing another one for some things, just to use a little feature,
then I'd make a similar feature to mean they didn't have to

[^Lele^]

Well, if the scene is created exactly with that intent, I see no problem for the different lighting. Maybe print a warning to the checkbox, similar to the additive mode in BlendMtl.

Not my choice to make.

Also, in the Corona image by fixeighted, the sun is still creating sharp and strong shadows. The size does not to have an effect on lighting that much.

I didn't say soft, i said divergent, like that of a spotlight. It's visible even in the image you mention. Open a demo of Corona and try for yourself.

As long as there is anything to control the size for the upper hemisphere of that squashed dome, I'd be happy...

You'll be in case able to repro on whatever you prefer, and tweak the geo accordingly.

Until you typed that last line I was just beginning to think of the vast amount of gifts that everyone would bestow upon you.

Well, i'll surely try and help with PoCs and early testing.

[matthew999]

Plus one for better implementation too and I'm sure there are many other vray users wanting the same.

[kosso_olli]

Just checking back to see if there are any improvements to this?