Realtime Radiosity

The two are not mutually exclusive, at least ray tracing doesn''t exclude radiosity.

If you truly want photorealistic then forget radiosity and go for a full GI solution, but the chances of that being realtime in the next 10 years I would predict are low (and I''d be more than happy to be proved wrong )

quote:

---

Original post by JuNC
The two are not mutually exclusive, at least ray tracing doesn't exclude radiosity.

---

Well sort of, but then you're no longer tracing rays to do the lighting, you're using radiosity to generate a lightmap, and then merely rendering the lightmap with raytracing. I'm talking more about the lighting then the actual rasterization.

EDIT: Yes, a full global ilumination renderer would be better, but radiosity is a portion of that, so forgetting it would make it rather difficult..


[edited by - cowsarenotevil on October 1, 2003 5:50:37 PM]

No, you can still use the raytracing to perform specular reflections and direct lighting, so you get the best of both worlds in that case.

EDIT:

The GI solution wouldn't be forgetting the radiosity now would it Since it is a *full* solution, including diffuse, directional diffuse, specular, anisotropic etc. etc. so it would automatically include radiosity under that umbrella. Of course you would no longer be using a radiosity technique and many of the optimizations that can bring, you need a different set of optimizations but get more 'realistic' images.



[edited by - JuNC on October 1, 2003 5:56:19 PM]

quote:

---

Original post by JuNC
No, you can still use the raytracing to perform specular reflections and direct lighting, so you get the best of both worlds in that case.

---

You say "no" as if you''re refuting one of my points. I''m now confused... But yes, that would be very cool...

Well you implied you wouldn''t use the raytracing for lighting and I was refuting that point

quote:

---

Original post by JuNC
Well you implied you wouldn''t use the raytracing for lighting and I was refuting that point

---

But you''re only using raytracing for part of the light... Just because you can combine them doesn''t mean you can''t compare them, otherwise the debates on neural nets and fuzzy logic couldn''t exist

But hang on, in your original post:

quote:

---

I decided to start this thread about how much realtime radiosity pwns raytracing

---

My contention was that they both constitute partial solutions and weren't mutually exclusive. So really what you just said kind of refuted your initial post

EDIT: We might be having a slight difference of terminology here, for most of the time I've been using 'radiosity' to mean 'radiosity techniques', mostly represented by Galerkin techniques and 'raytracing' to mean backward raytracing (ala Whitted).

[edited by - JuNC on October 1, 2003 6:23:18 PM]

This is the most useless arguement EVAR. But anyway, as I said, you can argue about weather neural nets are better than fuzzy logic, even though they can be combine, so why should these rendering techniques be different?

OK, to avoid this turning into a war of terms, heres my direct answers:

-It makes more realistic lighting

Agreed for indirect lighting, however you''ll never achieve good specular lighting with a pure radiosity technique (since it can by definition only deal with diffuse lighting)

-It doesn''t look to crispy (for lack of better word)

I prefer the term ''hyperrealistic'' and I totally agree. This might be reduced somewhat by using a more physically based colour model (i.e. spectral), but there are fundamental problems with energy conservation with the usual naive raytracing algorithm.

-It can be done on the vertex level

Hmm, I''m not sure this is a fair argument, I mean you could swing it round and ask how efficient is per-pixel radiosity?

quote:

---

Original post by JuNC
Agreed for indirect lighting, however you''ll never achieve good specular lighting with a pure radiosity technique (since it can by definition only deal with diffuse lighting)

---

Definately true. In fact, as far as I know, you can''t achieve specular lighting at all, because it''s dependant on the POV.

Exactly. Radiosity solutions and (backward) raytracing solutions complement each other nicely in that respect.

What I meant by that was that radiosity solutions provide a good time/quality tradeoff for indirect lighting. A more realistic full solution will take a lot of time, so mixing radiosity for indirect and raytracing (or even rasterizing with realistic BRDFs) for direct lighting provides some extremely high quality results.

[edited by - JuNC on October 1, 2003 7:00:42 PM]

photonmaps look bether than radiosity imho.. and solve the full lighting equation if you want at the same time, too..

---

If that''s not the help you''re after then you''re going to have to explain the problem better than what you have. - joanusdmentia

davepermen.net

>What I meant by that was that radiosity solutions
>provide a good time/quality tradeoff for indirect lighting.

>A more realistic full solution will take a lot of time, so >mixing radiosity for indirect and raytracing (or even

To clarify a bit: radiosity is good for indirect *diffuse* lighting.

Indirect non-diffuse (glossy, hightly specular) lighting is usually done using distrubution (ie. monte carlo) ray tracing.

Or at least that''s what they do with the Radiance program, as far as I remember. And as davepermen pointed out, Jensen''s photon maps do better (and are faster) than radiance (at least in Jensen''s papers *grin*).

Photon mapping switches adaptively between monte carlo ray tracing and photon map lookups according to the BRDF, btw. So it can handle slightly glossy surfaces nicely.

- Mikko Kauppila

PS. Threads like this make me understand why this section is called Graphics Programming and *Theory*

I found this an interesting read.

[edited by - dmounty on October 2, 2003 12:24:51 PM]

no pictures, no use.. i''ve seen some note on spherical harmonics in there.. bah..

i''ll read it later. it sounds at least like.. WOW.. so.. lets check how much real use it has..

---

If that''s not the help you''re after then you''re going to have to explain the problem better than what you have. - joanusdmentia

davepermen.net

okay. it has a realtime demo. much bether:D it looks nice, and while skimming trough the paper it sounds simple, and nice, too..

sofarsogood:D

---

If that''s not the help you''re after then you''re going to have to explain the problem better than what you have. - joanusdmentia

davepermen.net

yeah... it would have been wiser to post this link, as it contains the paper, the demo and a (270mb) video.

quote:

---

Original post by davepermen
photonmaps look bether than radiosity imho.. and solve the full lighting equation if you want at the same time, too...

---

Aside from photon maps having specular lighting and refraction, they look exactly the same, except photon maps are generally slower. And there may be ways of achieving specular lighting with radiosity (I have an idea, but I don''t know if it would work...)

quote:

---

Original post by uutee
To clarify a bit: radiosity is good for indirect *diffuse* lighting.

Indirect non-diffuse (glossy, hightly specular) lighting is usually done using distrubution (ie. monte carlo) ray tracing.

---

Of course it''s only a non-directional diffuse solution. I did allude to it earlier but I think I missed saying it explicitly in this thread. It still is a good time/quality tradeoff.

I actually currently use bi-directional path tracing for my off-line renderer, it is still a monte-carlo technique but it certainly isn''t distribution ray tracing. Looks pretty nice, although quite slow

Photon maps are really a way of accelerating (forward) path tracing, all of the techniques described here are just ways to approximate the full GI solution, each one capturing different aspects of it.