 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Just used one of my "never finished projects" to do a "quick" compare
between 3.6.1 and 3.7 beta30.rad1 radiosity with moderate low setting -
at least for this kind of scene. Lit by radiosity only, coming from a
skysphere with a high ambient value.
version 3.6:
global_settings {
assumed_gamma 1.0
max_trace_level 8
pretrace_start 0.08
pretrace_end 0.01
count 400
error_bound 0.75
nearest_count 5
low_error_factor 0.5
recursion_limit 3
}
version 3.7
global_settings {
max_trace_level 8
pretrace_start 16/image_width
pretrace_end 2/image_width
count 400
error_bound 0.75
nearest_count 1
low_error_factor 0.5
recursion_limit 3
}
I have also attached a version rendered by MCPov 0.5 with 1000 passes.
Well, look for yourself but from what I can tell the 3.7 radiosity
lighting looks quite strange. The floor is way too bright and other
objects like the rug on the table, the viol and the notes on the
virginal look totally flat. Note also this has nothing to do with the
change in gamma handling because I have always used assumed gamma of 1.0.
Here is a link to an image I have done back in 2002 with high quality
radiosity settings and (I guess) POV 3.5 at this time.
http://news.povray.org/povray.binaries.images/message/<3C3da07987%40news.povray.org>
There is a difference in the lighting because in the 2002 scene the 3rd
window is open but the 3 new test renders have this window draped.
Post a reply to this message
Attachments:
Download 'musiclesson pov 3.7 beta30 rad1.jpg' (115 KB)
Download 'musiclesson pov 3.6.1.jpg' (95 KB)
Download 'musiclesson mcpov 0.5 1000passes.jpg' (126 KB)
Preview of image 'musiclesson pov 3.7 beta30 rad1.jpg'
Preview of image 'musiclesson pov 3.6.1.jpg'
Preview of image 'musiclesson mcpov 0.5 1000passes.jpg'
|
|
| |
| |
|
|
|
|
| |
| |
|
|
That's too beautiful a scene to be just a test. ;) Inspired by the painting in
the lower right, I guess?
MCPov is the best really, specially with the sharp shadows. 3.6 seems to
conform with this lighting as well, except for the lack of sharp shadows. 3.7
does look too bright and a bit too flat on the table cloth and jar.
Did you use ambient 0 for all objects? BTW, I suppose 3.7 is the fastest,
right? How about tweaking it's settings a bit to see how it comes out with
more quality?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Ive <"ive### [at] lilysoft org"> wrote:
> Just used one of my "never finished projects" to do a "quick" compare
> between 3.6.1 and 3.7 beta30.rad1 radiosity with moderate low setting -
> at least for this kind of scene. Lit by radiosity only, coming from a
> skysphere with a high ambient value.
Impressive scene!
> Well, look for yourself but from what I can tell the 3.7 radiosity
> lighting looks quite strange. The floor is way too bright and other
> objects like the rug on the table, the viol and the notes on the
> virginal look totally flat. Note also this has nothing to do with the
> change in gamma handling because I have always used assumed gamma of 1.0.
Comments to the pictures:
- Render times?
- The MCPov shot looks most convincing, no argument here. However, it is not
perfectly clear whether this is due to an actual superiority in quality, or
imperfections in the POV 3.6/3.7 material setup. Especially a look at the
mirror raises suspicion: While it is rather dull in the MCPov shot, it seems to
be plain 100% reflective in the POV 3.6.1 and 3.7 shots. So there must be *some*
difference in material settings.
- The 3.6.1 shot looks rather dark; I attribute this to the windows and flaws in
POV 3.6 radiosity: When gathering deep-recursion samples, POV 3.6 effectively
just traced a single level, so I guess it will not make it through the windows.
All 3rd-bounce samples will probably be pitch black, and possibly even the
2nd-bounce ones.
- The 3.7 shot indeed looks weird - but this weirdness look distinctively
familiar: "Ambient" immediately comes to mind. It looks very much like you
forgot to set all materials to "ambient 0" (or, alternatively, turn down the
"ambient_light" to something like 0.001 and multiply your sky ambient by 1000;
usually this sufficiently dims the ambient terms and saves you a lot of work.)
> Here is a link to an image I have done back in 2002 with high quality
> radiosity settings and (I guess) POV 3.5 at this time.
>
> http://news.povray.org/povray.binaries.images/message/<3C3da07987%40news.povray.org>
That link only leads me to the p.b.i. overview.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
clipka wrote:
> Impressive scene!
>
thanks.
> Comments to the pictures:
>
> - Render times?
>
~2h for 3.6
~1h for 3.7
~20h for MCPov
I'm currently rendering a 3.7 version with higher quality settings and
will show it when it's finished.
> - The MCPov shot looks most convincing, no argument here. However, it is not
> perfectly clear whether this is due to an actual superiority in quality, or
> imperfections in the POV 3.6/3.7 material setup. Especially a look at the
> mirror raises suspicion: While it is rather dull in the MCPov shot, it seems to
> be plain 100% reflective in the POV 3.6.1 and 3.7 shots. So there must be *some*
> difference in material settings.
>
I have no explanation why the reflection in the mirror looks different
in the MCPov version. And in fact it makes me quite unhappy, it looks
somewhat desaturated. But believe me, the material used for the mirror
is in all shots exactly the same. The only difference in the materials
are the woods for the chairs, the viol and the mirror frame where the
MCPov version makes use of the MCPov blurred reflection feature where
the 3.6/3.7 versions use averaged normals to get some highlights.
> - The 3.6.1 shot looks rather dark; I attribute this to the windows and flaws in
> POV 3.6 radiosity: When gathering deep-recursion samples, POV 3.6 effectively
> just traced a single level, so I guess it will not make it through the windows.
> All 3rd-bounce samples will probably be pitch black, and possibly even the
> 2nd-bounce ones.
>
In fact there is NO window "glass", there is just empty space ;)
> - The 3.7 shot indeed looks weird - but this weirdness look distinctively
> familiar: "Ambient" immediately comes to mind. It looks very much like you
> forgot to set all materials to "ambient 0" (or, alternatively, turn down the
> "ambient_light" to something like 0.001 and multiply your sky ambient by 1000;
> usually this sufficiently dims the ambient terms and saves you a lot of work.)
>
Nope. All materials do use ambient 0 in the finish statement and no
default include files are used where something else might be specified.
> That link only leads me to the p.b.i. overview.
>
Sorry. This should work:
http://news.povray.org/povray.binaries.images/thread/<3da07987%40news.povray.org>/?ttop=296700&toff=5700
-Ive
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
nemesis wrote:
> That's too beautiful a scene to be just a test. ;) Inspired by the painting in
> the lower right, I guess?
"The Music Lesson" by Vermeer
> MCPov is the best really, specially with the sharp shadows. 3.6 seems to
> conform with this lighting as well, except for the lack of sharp shadows. 3.7
> does look too bright and a bit too flat on the table cloth and jar.
>
> Did you use ambient 0 for all objects?
Sure.
> BTW, I suppose 3.7 is the fastest, right?
yes.
> How about tweaking it's settings a bit to see how it comes out with
> more quality?
>
Doing this right now ;)
-Ive
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Ive <"ive### [at] lilysoftorg"> wrote:
> I have no explanation why the reflection in the mirror looks different
> in the MCPov version. And in fact it makes me quite unhappy, it looks
> somewhat desaturated.
Hum... maybe max trace level issues of sorts...?
> > - The 3.6.1 shot looks rather dark; I attribute this to the windows and flaws in
> > POV 3.6 radiosity: When gathering deep-recursion samples, POV 3.6 effectively
> > just traced a single level, so I guess it will not make it through the windows.
> > All 3rd-bounce samples will probably be pitch black, and possibly even the
> > 2nd-bounce ones.
> >
>
> In fact there is NO window "glass", there is just empty space ;)
Hum again... okay, so there is no spoon...
If the scene is just radiosity-lit, and materials are defined deliberately to be
identical, then the only thing I can think of that could cause a significant
bias to any particular color (like, for example, black) is POV-Ray giving up on
some rays due to max trace level, ADC bailout or similar (i.e. recursion depth
in case of radiosity) - in which cases POV will invariably use pitch black as
substitute.
So the way a version of POV "counts" recursion depth might make a difference
regarding how much black is mixed in.
Does a significant portion of surfaces in your shot have reflective components?
> > - The 3.7 shot indeed looks weird - but this weirdness look distinctively
> > familiar: "Ambient" immediately comes to mind.
>
> Nope. All materials do use ambient 0 in the finish statement and no
> default include files are used where something else might be specified.
That's perfectly weird. In all tests shots, I have never seen such effects. Note
that the white tiles do not only "bleed" onto the walls - they themselves look
overly bright.
Something gets a >1 term into it somewhere.
Again: Does a significant portion of surfaces in your shot have reflective
components?
Assuming for now that the tiles are partially reflective - is the sum of their
"reflection" and "diffuse" values possibly >1?
Note that this breaks the law of conservation of energy. This is not an issue in
classically-lit POV scenes, because of the strict separation between "images"
and "light sources"; however, in radiosity scenes it can become dramatic.
I can imagine that MCPov *might* handle such situations automatically as a side
effect of its design. However, I'm quite sure standard POV does not.
Just to make sure: We are talking about 3.7.0.beta.30-rad1, right?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
clipka wrote:
> Ive <"ive### [at] lilysoftorg"> wrote:
>> I have no explanation why the reflection in the mirror looks different
>> in the MCPov version. And in fact it makes me quite unhappy, it looks
>> somewhat desaturated.
>
> Hum... maybe max trace level issues of sorts...?
>
Hmm, maybe. But shouldn't be max_trace_level of 8 be enough? It's just
camera -> mirror -> rug. And the rug is not reflective.
>>> - The 3.6.1 shot looks rather dark; I attribute this to the windows and flaws in
>>> POV 3.6 radiosity: When gathering deep-recursion samples, POV 3.6 effectively
>>> just traced a single level, so I guess it will not make it through the windows.
>>> All 3rd-bounce samples will probably be pitch black, and possibly even the
>>> 2nd-bounce ones.
>>>
>> In fact there is NO window "glass", there is just empty space ;)
>
> Hum again... okay, so there is no spoon...
>
> If the scene is just radiosity-lit, and materials are defined deliberately to be
> identical, then the only thing I can think of that could cause a significant
> bias to any particular color (like, for example, black) is POV-Ray giving up on
> some rays due to max trace level, ADC bailout or similar (i.e. recursion depth
> in case of radiosity) - in which cases POV will invariably use pitch black as
> substitute.
>
> So the way a version of POV "counts" recursion depth might make a difference
> regarding how much black is mixed in.
>
> Does a significant portion of surfaces in your shot have reflective components?
>
All woods have a small amount of blurred reflection to get highlights.
In the original 2002 version also the floor tiles where slightly
reflective but I have disabled the floor reflection for this test
render. The original version also used more complicated layered textures
for the walls, and this new renders do use simplified texture there.
The layered textures caused some strange artifacts in the 2002 render
and where the main reason I gave up at this time.
>
>>> - The 3.7 shot indeed looks weird - but this weirdness look distinctively
>>> familiar: "Ambient" immediately comes to mind.
>> Nope. All materials do use ambient 0 in the finish statement and no
>> default include files are used where something else might be specified.
>
> That's perfectly weird. In all tests shots, I have never seen such effects. Note
> that the white tiles do not only "bleed" onto the walls - they themselves look
> overly bright.
>
> Something gets a >1 term into it somewhere.
>
> Again: Does a significant portion of surfaces in your shot have reflective
> components?
>
> Assuming for now that the tiles are partially reflective - is the sum of their
> "reflection" and "diffuse" values possibly >1?
Nope. I would never do this ;) (And I have always double checked in the
SDL source files to really make sure my answers and description is
correct.) The color AND diffuse values are alwayxs <= 1. And, as
mentioned, for the new renders NO reflection for the floor.
> Note that this breaks the law of conservation of energy. This is not an issue in
> classically-lit POV scenes, because of the strict separation between "images"
> and "light sources"; however, in radiosity scenes it can become dramatic.
>
I know, I really do.
> I can imagine that MCPov *might* handle such situations automatically as a side
> effect of its design. However, I'm quite sure standard POV does not.
>
>
> Just to make sure: We are talking about 3.7.0.beta.30-rad1, right?
>
Yes, 3.7.0.beta.30-rad1 from Chris' zip file, exactly.
I have attached a version with higher quality settings (and about 5h
render time):
radiosity {
pretrace_start 8/image_width
pretrace_end 2/image_width
count 800
error_bound 0.4
nearest_count 1
low_error_factor 0.5
recursion_limit 3
brightness 1
}
And also a version where the radiosity block is just commented out (and
no other changes are made). As expected, everything is pitch black
except the sky and reflective parts of the scene where the sky gets
reflected.
I really hate it to say so, but I think there is something weird going
on with the radiosity calculation.
-Ive
Post a reply to this message
Attachments:
Download 'musiclesson 3.7beta30 - 2.jpg' (89 KB)
Download 'musiclesson- nolit.jpg' (11 KB)
Preview of image 'musiclesson 3.7beta30 - 2.jpg'
Preview of image 'musiclesson- nolit.jpg'
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Ive <"ive### [at] lilysoftorg"> wrote:
> >> I have no explanation why the reflection in the mirror looks different
> >> in the MCPov version. And in fact it makes me quite unhappy, it looks
> >> somewhat desaturated.
> >
> > Hum... maybe max trace level issues of sorts...?
>
> Hmm, maybe. But shouldn't be max_trace_level of 8 be enough? It's just
> camera -> mirror -> rug. And the rug is not reflective.
Well, I don't know how MCPov handles trace level when it comes to diffuse
bounces. 'Twas just an idea anyway.
> All woods have a small amount of blurred reflection to get highlights.
The woods are not the "bad guys" - the tiles are, I think.
> I have attached a version with higher quality settings (and about 5h
> render time):
Didn't compare the shots in detail, but one thing is for sure: The weird effect
is still in there.
> And also a version where the radiosity block is just commented out (and
> no other changes are made). As expected, everything is pitch black
> except the sky and reflective parts of the scene where the sky gets
> reflected.
>
> I really hate it to say so, but I think there is something weird going
> on with the radiosity calculation.
Obviously.
Would you mind e-mailing the scene to "christoph (at) lipka-koeln (dot) de"? As
of now, it's the only scene I've seen so far that shows this effect.
Maybe you can strip it down to just the room and the tiles.
Another thought that just crossed my mind is "gamma": what if 3.7 interprets the
sky sphere differently with respect to gamma? (Then again, that would probably
look much different.)
At the moment I'm suspecting the floor to pick up light from below for some
reason. What is the floor made of, geometrically speaking?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
clipka wrote:
> Would you mind e-mailing the scene to "christoph (at) lipka-koeln (dot) de"? As
> of now, it's the only scene I've seen so far that shows this effect.
>
> Maybe you can strip it down to just the room and the tiles.
>
Will do so, but may take a little time.
> Another thought that just crossed my mind is "gamma": what if 3.7 interprets the
> sky sphere differently with respect to gamma? (Then again, that would probably
> look much different.)
>
> At the moment I'm suspecting the floor to pick up light from below for some
> reason. What is the floor made of, geometrically speaking?
>
I will also do some experiments to see if I can narrow down the problem.
-Ive
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
clipka wrote:
> At the moment I'm suspecting the floor to pick up light from below for some
> reason. What is the floor made of, geometrically speaking?
>
Just a box ;)
#declare W = 500.0;
#declare L = 664.6;
#declare H = 323.6;
//==============================================================
// room
//==============================================================
union {
// floor
box { <0,0, 1> <W,-1,-L-1> texture {T_Floor} }
... and a lot of more stuff
But here is the interesting thing:
#declare T_Floor = texture {
material_map { png "Floor#2"
texture {T_FloorMortar}
texture {T_FloorWhite}
texture {T_FloorBlack}
texture {T_FloorBlack translate 79}
texture {T_FloorBlack translate 255}
}
rotate x*90
rotate y*45
scale 117.2/2
translate <-20.0, 0, 0>
}
where each of the separate textures is a quite complicated layered
texture. Suspicion was that either the heavy layered textures or the
material_map statement did mess up things.
So a quick simplification:
#declare T_FloorWhite_simple = texture
{
pigment {rgb 0.55}
finish {ambient 0 diffuse 1}
}
#declare T_FloorBlack_simple = texture
{
pigment {rgb 0.05}
finish {ambient 0 diffuse 1}
}
#declare T_Floor = texture {
checker
texture {T_FloorWhite_simple}, texture {T_FloorWhite_simple}
rotate x*90
rotate y*45
scale 117.2/2
translate <-20.0, 0, 0>
}
and the result is still weird! See picture 1.
But then:
#declare T_Floor = texture {
texture {T_FloorWhite_simple}
rotate x*90
rotate y*45
scale 117.2/2
translate <-20.0, 0, 0>
}
works perfectly as expected (low quality aside). Picture 2.
This effect can be easily reproduced in any other scene.
So it seems to me that the problem is maybe not even related to
radiosity but it is to patterned textures. Maybe it has something to do
with he post by me to p.beta-test from 4/25/2008 "texture_map vs.
pigment_map" - that never got a response ;(
-Ive
Post a reply to this message
Attachments:
Download 'musiclesson-simple1.jpg' (50 KB)
Download 'musiclesson-simple2.jpg' (77 KB)
Preview of image 'musiclesson-simple1.jpg'
Preview of image 'musiclesson-simple2.jpg'
|
|
| |
| |
|
|
|
|
| |
|
|
