> No surprise at all. Path tracing works very well with large light sources. And
> the largest one possible is usually sky. Most rays terminate very quickly and
> the image converges very fast.
I agree with the theory. But the patio (a previuos screne that I posted in this
thread) was rendered with the same lighting (a portal around a point light and
a sky dome) and the rendering was far slower.

Regards,
Fidos

Post a reply to this message

> See my thread in OT. Especially my last post where I realised that this works.
> There is some wrong info also in the thread. My last post should be correct.
Thank you for this precise information.
>
> Bidirectional rendering looks interesting too.
I agree. It seems to have a good potential. The solve of caustic should be
easier. The difficulty seems to weight correctly the 2 paths when merged.

Regards,
Fidos

Post a reply to this message

In order to test a new way to sample the portal, I rendered the kitchen scene of
Jaime Vives Piqueres in order to compare with a previous post
(http://news.povray.org/web.47b1f7addcec6a9f3a81b2650%40news.povray.org).

Because I reduced the max trace level, the conclusion in not direct.

But the spots and the noise reduced for half the time (12h instead of 24h).

Regards,
Fidos.

Post a reply to this message

Attachments:
Download '2nd-pass.05.jpg' (379 KB)

Preview of image '2nd-pass.05.jpg'

>> No surprise at all. Path tracing works very well with large light sources. And
>> the largest one possible is usually sky. Most rays terminate very quickly and
>> the image converges very fast.
> I agree with the theory. But the patio (a previuos screne that I posted in this
> thread) was rendered with the same lighting (a portal around a point light and
> a sky dome) and the rendering was far slower.

   It must be because Gilles scene is mostly triangle meshes, while my patio 
scene is mostly inefficient isosurfaces.

--
Jaime

Post a reply to this message

My newest image. Now with specular reflections and infinite recursion
level. (Russian roulette style termination.) This was about a 10minute
render. A large overcast light makes the noise go away pretty fast. I so
love those natural shadows under the spheres.


Next I restructure the code to allow easy addition of different kinds of
objects. Still stuck at spheres. Also refractions are in the list.

Post a reply to this message

Attachments:
Download 'kuva6.jpg' (43 KB)

Preview of image 'kuva6.jpg'

> Next I restructure the code to allow easy addition of different kinds of
> objects. Still stuck at spheres. Also refractions are in the list.
> 

Ah, yours is written from scratch? If you were modifying POV-Ray, it 
would already support everything from spheres to isosurfaces :)

Post a reply to this message

Severi Salminen <sev### [at] NOTTHISsaunalahtifiinvalid> wrote:
> My newest image. Now with specular reflections and infinite recursion
> level. (Russian roulette style termination.) This was about a 10minute
> render. A large overcast light makes the noise go away pretty fast. I so
> love those natural shadows under the spheres.
Good picture. The blending of the sphere colors on the ground is nice.

Regards,
Fidos

Post a reply to this message

Nicolas Alvarez wrote:
>> Next I restructure the code to allow easy addition of different kinds of
>> objects. Still stuck at spheres. Also refractions are in the list.
> 
> Ah, yours is written from scratch? If you were modifying POV-Ray, it
> would already support everything from spheres to isosurfaces :)

Nah, it is a lot more fun to do everything by yourself :) This project
actually started because I didn't know C++ (nor Linux X programming) at
all. Then i saw this thread and got really interested in path tracing.

And I love the results so far: it is so easy to make more realistic
looking images with a path tracer than with "conventional" ray tracer
IMO. Even the last sphere image I posted looks so natural to me. I
wonder how easy it would be to reproduce it with POV.

Post a reply to this message

Severi Salminen napsal(a):
> Nicolas Alvarez wrote:
>>> Next I restructure the code to allow easy addition of different kinds of
>>> objects. Still stuck at spheres. Also refractions are in the list.
>> Ah, yours is written from scratch? If you were modifying POV-Ray, it
>> would already support everything from spheres to isosurfaces :)
> 
> Nah, it is a lot more fun to do everything by yourself :) This project
> actually started because I didn't know C++ (nor Linux X programming) at
> all. Then i saw this thread and got really interested in path tracing.
> 
> And I love the results so far: it is so easy to make more realistic
> looking images with a path tracer than with "conventional" ray tracer
> IMO. Even the last sphere image I posted looks so natural to me. I
> wonder how easy it would be to reproduce it with POV.
using high radiosity settings with no conventional light source will do 
the same. I'm thinking of:
radiosity{
   error_bound 0        //to prevent any reuse
   count 1              //to prevent getting damn slow
   pretrace_start 1     //
   pretrace_end 1       //these won't be used either.
   minimun_reuse 0      //necessary
   normal on	       //this won't hurt the performance
   #declare s=seed(Parse_String(date("%Y%y%j%H%M%S")))
   #declare d=VRand_On_Sphere(s);
   #if(d.y<0)
     d=-d;
   #end
   samples{1,d}
   randomize on         //a (lame) attempt to randomize the sample space.
                        //adapt adc_bailout and max_recursion as you like
}
The problem is that the light ray always reaches the surface at the same 
angle (different from ray to ray). This might (and will) produce 
artifacts (honestly, I'm interested in these artifacts), it might even 
go all wrong (the first bunce will be OK).

-- 
You know you've been raytracing too long when...
you start thinking up your own "You know you've been raytracing too long 
when..." sigs (I did).
-Johnny D

Post a reply to this message

> Next I restructure the code to allow easy addition of different kinds of
> objects. Still stuck at spheres. Also refractions are in the list.

And here are the planes, finally! Had to learn to do abstract classes in
C++. Now I can easily add different kinds of objects and I have to only
specify a few functions specific to that new object type. The basic
rendering code don't need to know anything about the object. Late
binding takes care of the dirty details of calling the correct method.
It introduces a bit of overhead but the code remains very simple. I
consider that an advantage.

This image is the same as before but now inside a room made of 6 planes.
Because of enclosed space the rays keep on bouncing and bouncing. So it
takes time for the noise to go away.

I also modified my anti-aliasing method. I used to divide the pixel into
equal subpixels, trace them, and average the results. No need to do it
anymore. I just shoot a random ray anywhere through the pixel being
rendered. As the final image still goes through hundreds/thousands of
passes you get pretty much perfect AA. (The attached image is still made
with the old method, though...)

Post a reply to this message

Attachments:
Download 'kuva7.jpg' (34 KB)

Preview of image 'kuva7.jpg'