Here is the result of trying to visually match a Megapov 0.5 scene
with a CCD image of the Cornell box. The only assumption made, based
on observing the image on a calibrated monitor, was that the CCD image
gamma is 2.2. This gamma was applied to the final image and extracted
colors. Some red and green was subtracted from the extracted light
surface color to remove the effects of color bleeding. Extracted red
and green colors were only multiplied by a factor in linear space to
match intensity. Area light color is RGB white. The POV-Ray scene
geometry is reconstructed from the CCD image so it doesn't match
exactly with the physical box. I think it's good enough for artistic
purposes and considering that POV wasn't designed to produce realistic
simulations :)

The most difficult part was to get the darkening effect at the corner
of the back wall and ceiling and it's not even that good. This also
produced an anomaly at the small box/floor intersection. Notice also
the good red color bleed on the tall box and not so good green color
bleed on the small box/floor. The brightness of the lower front face
of the tall box could be caused by different scene geometry or the
incorrect light distribution from the area light (It's really a point
light in all but the shadow calculations. This is why large area lights
don't produce correct direct illumination. A patch to fix this would
be nice :). Spotlight was used to simulate the cosine intensity falloff
from planar light source.

I'm not sure whether the settings I used are generally valid for all kinds
of scenes or just this one. I will post the source with comments later
so that the settings can be tested on different scenes and conditions.

The most notable changes from the MP recommended values are:
  brightness 2.5 
  gray_threshold 0.5


First image: CCD image of the Cornell box.

Second image: Rendering time about 0.5h but reduces significantly
              without the (minor) corner effect. Requires more tweaking
              to reduce rendering time.

Third image: Megapov 0.5 high quality radiosity settings. 1.5min.

Last: No radiosity. 20sec


-----------------------------------------------------------------------
Kari Kivisalo                                          www.kivisalo.net

Post a reply to this message

Attachments:
Download 'cornell2.jpg' (55 KB)

Preview of image 'cornell2.jpg'

I would say this is typical; the official POV is much slower but much more
realistic...

--
David Fontaine     <dav### [at] faricynet>     ICQ 55354965
Please visit my website: http://www.faricy.net/~davidf/

Post a reply to this message

David Fontaine wrote:
> 
> I would say this is typical; the official POV is much slower but much more
> realistic...

All 3 pics done in megapov 0.5.

-----------------------------------------------------------------------
Kari Kivisalo                                          www.kivisalo.net

Post a reply to this message

Kari Kivisalo wrote:

> All 3 pics done in megapov 0.5.

Oh! 3 is a different method?

--
David Fontaine     <dav### [at] faricynet>     ICQ 55354965
Please visit my website: http://www.faricy.net/~davidf/

Post a reply to this message

Kari Kivisalo wrote:
> 
> Here is the result of trying to visually match a Megapov 0.5 scene
> with a CCD image of the Cornell box. The only assumption made, based
> on observing the image on a calibrated monitor, was that the CCD image
> gamma is 2.2. This gamma was applied to the final image and extracted
> colors. Some red and green was subtracted from the extracted light
> surface color to remove the effects of color bleeding. Extracted red
> and green colors were only multiplied by a factor in linear space to
> match intensity. Area light color is RGB white. The POV-Ray scene
> geometry is reconstructed from the CCD image so it doesn't match
> exactly with the physical box. I think it's good enough for artistic
> purposes and considering that POV wasn't designed to produce realistic
> simulations :)
> 
> The most difficult part was to get the darkening effect at the corner
> of the back wall and ceiling and it's not even that good. This also
> produced an anomaly at the small box/floor intersection. Notice also
> the good red color bleed on the tall box and not so good green color
> bleed on the small box/floor. The brightness of the lower front face
> of the tall box could be caused by different scene geometry or the
> incorrect light distribution from the area light (It's really a point
> light in all but the shadow calculations. This is why large area lights
> don't produce correct direct illumination. A patch to fix this would
> be nice :). Spotlight was used to simulate the cosine intensity falloff
> from planar light source.
> 
> I'm not sure whether the settings I used are generally valid for all kinds
> of scenes or just this one. I will post the source with comments later
> so that the settings can be tested on different scenes and conditions.
> 
> The most notable changes from the MP recommended values are:
>   brightness 2.5
>   gray_threshold 0.5
> 
> First image: CCD image of the Cornell box.
> 
> Second image: Rendering time about 0.5h but reduces significantly
>               without the (minor) corner effect. Requires more tweaking
>               to reduce rendering time.
> 
> Third image: Megapov 0.5 high quality radiosity settings. 1.5min.
> 
> Last: No radiosity. 20sec

Looking at the first image and comparing it with the third, I'm somewhat
curious what would happen if you played with the radiosity brightness.

-- 
Xplo Eristotle
http://start.at/xplosion/

"And then one day you find ten years have got behind you
No one told you when to run, you missed the starting gun"
    -Pink Floyd

Post a reply to this message

Does CCD mean a photograph of an actual object?

Kari Kivisalo wrote:

> Here is the result of trying to visually match a Megapov 0.5 scene
> with a CCD image of the Cornell box. The only assumption made, based
> on observing the image on a calibrated monitor, was that the CCD image
> gamma is 2.2. This gamma was applied to the final image and extracted
> colors. Some red and green was subtracted from the extracted light
> surface color to remove the effects of color bleeding. Extracted red
> and green colors were only multiplied by a factor in linear space to
> match intensity. Area light color is RGB white. The POV-Ray scene
> geometry is reconstructed from the CCD image so it doesn't match
> exactly with the physical box. I think it's good enough for artistic
> purposes and considering that POV wasn't designed to produce realistic
> simulations :)
>
> The most difficult part was to get the darkening effect at the corner
> of the back wall and ceiling and it's not even that good. This also
> produced an anomaly at the small box/floor intersection. Notice also
> the good red color bleed on the tall box and not so good green color
> bleed on the small box/floor. The brightness of the lower front face
> of the tall box could be caused by different scene geometry or the
> incorrect light distribution from the area light (It's really a point
> light in all but the shadow calculations. This is why large area lights
> don't produce correct direct illumination. A patch to fix this would
> be nice :). Spotlight was used to simulate the cosine intensity falloff
> from planar light source.
>
> I'm not sure whether the settings I used are generally valid for all kinds
> of scenes or just this one. I will post the source with comments later
> so that the settings can be tested on different scenes and conditions.
>
> The most notable changes from the MP recommended values are:
>   brightness 2.5
>   gray_threshold 0.5
>
> First image: CCD image of the Cornell box.
>
> Second image: Rendering time about 0.5h but reduces significantly
>               without the (minor) corner effect. Requires more tweaking
>               to reduce rendering time.
>
> Third image: Megapov 0.5 high quality radiosity settings. 1.5min.
>
> Last: No radiosity. 20sec
>
> -----------------------------------------------------------------------
> Kari Kivisalo                                          www.kivisalo.net
>
>   ------------------------------------------------------------------------
>  [Image]

Post a reply to this message

"Greg M. Johnson" wrote:
> 
> Does CCD mean a photograph of an actual object?
> 

Glen Berry gave the link for the Cornell Box. It's a real 
exisiting box in p.u.p.

   http://www.graphics.cornell.edu/online/box/

And you are right about the photograph.

Marc

-- 
Marc Schimmler

Post a reply to this message

On Wed, 31 May 2000 07:22:57 -0400, "Greg M. Johnson"
<gre### [at] my-dejanewscom> wrote:

>Does CCD mean a photograph of an actual object?

In this case, yes. For the record, "CCD" is an abreviation for "Charge
Coupled Device." CCD sensors are used in digital cameras. Cornell
University used a precision digital camera to photogragh an actual
box, "The Cornell Box", and used this as a reference for comparing
computer rendering systems. 

They also compensated for spectral response of the CCD sensor, the
spectral response of the camera lens, the spatial distortion of the
camera lens, polarization effects, and several other factors. They
also broke the visible spectrum into 76 seperate wavelength groups,
compared to POV-Ray, which only seperates light into the three groups
of Red Green and Blue. It would be *extremely* difficult for an
individual to make an exact replica of their testing system.

Still, I am considering constructing a simplified version of the
Cornell Box, or something similar. I'm thinking of building a physical
box, photographing it, and using that for radiosity comparisons. I'll
let eveyone know what sort of progress I achieve, if I actually go
ahead with the project. My main concern is making the project
something that another POV enthusiast could duplicate, unlike the
Cornell Box, which requires laboratory-grade equipment to deal with. 

Even though my tests would be much more crude than Cornell's, I think
they would still provide some insight for fellow raytracers. I
encourage anyone else interested in this sort of physical testing to
contact me through personal email or this news server.

By the way, which newsgroup should this project be posted in?

   povray.general  ?  
   povray.advanced-users ??

Later,
Glen Berry

( Remove the "7" from 7no### [at] ezwvcom to email me. )

Post a reply to this message