 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Hi all - I ended up with a little bit of time, so I thought I'd try combining
some of my techniques into a test render to see what sort of quality I could
get.
This is a high quality mesh of a Buddha (from Jotero.com I think) with one
(approx) 40x40x40 DF3 proximity pattern applied to three textures (bronze, stone
and steel).
The environment map is an HDR lightprobe I did of my old office (with a nice
balance of outside and inside lighting).
The lighting rig consists of 64 area lights generated by LighMapGen. I set the
"no_radiosity" flag on the environment map, and then turned on a very low
quality radiosity pass (20 samples) with a recursion limit of 1.
The scene was then rendered in multiple passes using animation and my Camera35mm
macros running in "stochastic rendering mode", which gives rise to the
anti-aliasing, focal blur and blurred reflections (on the stone).
Each pass of the 1280x720 scene took approx 2 minutes, and I let it run over
night, and in the morning combined about 300 passes into the final image.
Everything was rendered on my 2.26MHz MacBook Pro in the 64-bit bui;d of POV-Ray
3.7 b36.
I'll also post a comparison shot of very high quality radiosity on the same
scene, which gives almost identical lighting in it's 7 hour render, but without
anti-aliasing, focal blur or blurred reflections.
The main area I'm unhappy with is the focal blur - it's not as smooth as I would
like. I know another few hundred passes would fix that, but I think I've thought
up a great way to use the user_defined camera functions to end up with better
quality in about 1/4 the time. It's a shame I don't have them in POR-Ray 3.7!
Comments? Questions?
Cheers,
Edouard.
Post a reply to this message
Attachments:
Download '3-buddhas-multipass.jpg' (249 KB)
Preview of image '3-buddhas-multipass.jpg'
|
|
| |
| |
|
|
|
|
| |
| |
|
|
As a comparison, here is the same scene done with radiosity. I used a sample
count of 800 and a recursion limit of 2. The render took about 7 hours.
I tried reusing the radiosity data from the cache it created for another render,
but the render time didn't come down dramatically (a bit over two hours for the
second render I think).
Cheers,
Edouard.
Post a reply to this message
Attachments:
Download '3-buddhas-radiosity-hq.jpg' (283 KB)
Preview of image '3-buddhas-radiosity-hq.jpg'
|
|
| |
| |
|
|
|
|
| |
| |
|
|
And finally a graphic showing how the image quality improves over time with
multipass rendering. I reckon I could have stopped at two hours if it weren't
for the artifacts in the out of focus areas.
Cheers,
Edouard.
Post a reply to this message
Attachments:
Download 'multipass-3.7.jpg' (709 KB)
Preview of image 'multipass-3.7.jpg'
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Edouard" <pov### [at] edouard info> wrote:
> Comments? Questions?
Woweee.
Definitely a photo. Arrest that man!
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Stunningly realistic materials & rendering techniques. I genuinely thought
it was a photo.
I'm half tempted to out-do you by printing a piece of paper with a grid of
measurements & pov logos and taking a photo of an ornament, to demonstrate
*total* photo realism! Mwahahaha!
I'd had a similar thought about user defined cameras and stochastic
sampling, primarily because pov's apeture shape for focal blur is a cube,
which is unlike any camera that could ever be built! But I gave up when the
calculations for a nice random distribution on a 5-blade iris started to
hurt my brain. If you crack it (the problem, not my brain) I'd love to see
the code!
--
Tek
http://evilsuperbrain.com
"Edouard" <pov### [at] edouardinfo> wrote in message
news:web.4bda380478961eac21619a220@news.povray.org...
> Hi all - I ended up with a little bit of time, so I thought I'd try
> combining
> some of my techniques into a test render to see what sort of quality I
> could
> get.
>
> This is a high quality mesh of a Buddha (from Jotero.com I think) with one
> (approx) 40x40x40 DF3 proximity pattern applied to three textures (bronze,
> stone
> and steel).
>
> The environment map is an HDR lightprobe I did of my old office (with a
> nice
> balance of outside and inside lighting).
>
> The lighting rig consists of 64 area lights generated by LighMapGen. I set
> the
> "no_radiosity" flag on the environment map, and then turned on a very low
> quality radiosity pass (20 samples) with a recursion limit of 1.
>
> The scene was then rendered in multiple passes using animation and my
> Camera35mm
> macros running in "stochastic rendering mode", which gives rise to the
> anti-aliasing, focal blur and blurred reflections (on the stone).
>
> Each pass of the 1280x720 scene took approx 2 minutes, and I let it run
> over
> night, and in the morning combined about 300 passes into the final image.
>
> Everything was rendered on my 2.26MHz MacBook Pro in the 64-bit bui;d of
> POV-Ray
> 3.7 b36.
>
> I'll also post a comparison shot of very high quality radiosity on the
> same
> scene, which gives almost identical lighting in it's 7 hour render, but
> without
> anti-aliasing, focal blur or blurred reflections.
>
> The main area I'm unhappy with is the focal blur - it's not as smooth as I
> would
> like. I know another few hundred passes would fix that, but I think I've
> thought
> up a great way to use the user_defined camera functions to end up with
> better
> quality in about 1/4 the time. It's a shame I don't have them in POR-Ray
> 3.7!
>
> Comments? Questions?
>
> Cheers,
> Edouard.
>
Post a reply to this message
|
|
| |
| |
|
|
From: Tek
Subject: Re: Three Buddhas - Multipass Rendering vs Radiosity
Date: 30 Apr 2010 19:45:39
Message: <4bdb6ba3@news.povray.org>
|
|
|
| |
| |
|
|
BTW is there any chance you could show us the source to those materials?
I'd love to know if they're incredibly precisely based on reality or if they
just look real because of the proximity pattern and how you've lit &
"photographed" them.
--
Tek
http://evilsuperbrain.com
"Tek" <tek### [at] evilsuperbraincom> wrote in message
news:4bdb6aea$1@news.povray.org...
> Stunningly realistic materials & rendering techniques. I genuinely thought
> it was a photo.
>
> I'm half tempted to out-do you by printing a piece of paper with a grid of
> measurements & pov logos and taking a photo of an ornament, to demonstrate
> *total* photo realism! Mwahahaha!
>
> I'd had a similar thought about user defined cameras and stochastic
> sampling, primarily because pov's apeture shape for focal blur is a cube,
> which is unlike any camera that could ever be built! But I gave up when
> the calculations for a nice random distribution on a 5-blade iris started
> to hurt my brain. If you crack it (the problem, not my brain) I'd love to
> see the code!
>
> --
> Tek
> http://evilsuperbrain.com
>
> "Edouard" <pov### [at] edouardinfo> wrote in message
> news:web.4bda380478961eac21619a220@news.povray.org...
>> Hi all - I ended up with a little bit of time, so I thought I'd try
>> combining
>> some of my techniques into a test render to see what sort of quality I
>> could
>> get.
>>
>> This is a high quality mesh of a Buddha (from Jotero.com I think) with
>> one
>> (approx) 40x40x40 DF3 proximity pattern applied to three textures
>> (bronze, stone
>> and steel).
>>
>> The environment map is an HDR lightprobe I did of my old office (with a
>> nice
>> balance of outside and inside lighting).
>>
>> The lighting rig consists of 64 area lights generated by LighMapGen. I
>> set the
>> "no_radiosity" flag on the environment map, and then turned on a very low
>> quality radiosity pass (20 samples) with a recursion limit of 1.
>>
>> The scene was then rendered in multiple passes using animation and my
>> Camera35mm
>> macros running in "stochastic rendering mode", which gives rise to the
>> anti-aliasing, focal blur and blurred reflections (on the stone).
>>
>> Each pass of the 1280x720 scene took approx 2 minutes, and I let it run
>> over
>> night, and in the morning combined about 300 passes into the final image.
>>
>> Everything was rendered on my 2.26MHz MacBook Pro in the 64-bit bui;d of
>> POV-Ray
>> 3.7 b36.
>>
>> I'll also post a comparison shot of very high quality radiosity on the
>> same
>> scene, which gives almost identical lighting in it's 7 hour render, but
>> without
>> anti-aliasing, focal blur or blurred reflections.
>>
>> The main area I'm unhappy with is the focal blur - it's not as smooth as
>> I would
>> like. I know another few hundred passes would fix that, but I think I've
>> thought
>> up a great way to use the user_defined camera functions to end up with
>> better
>> quality in about 1/4 the time. It's a shame I don't have them in POR-Ray
>> 3.7!
>>
>> Comments? Questions?
>>
>> Cheers,
>> Edouard.
>>
>
>
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Tek" <tek### [at] evilsuperbraincom> wrote:
> Stunningly realistic materials & rendering techniques. I genuinely thought
> it was a photo.
Thank-you so much.
> I'd had a similar thought about user defined cameras and stochastic
> sampling, primarily because pov's apeture shape for focal blur is a cube,
> which is unlike any camera that could ever be built! But I gave up when the
> calculations for a nice random distribution on a 5-blade iris started to
> hurt my brain. If you crack it (the problem, not my brain) I'd love to see
> the code!
Yes - POVs aperture code is pretty substandard.
I just use a pigment or a PNG image of the bokeh, and sample it with an halton
sequence. I then dim the image by the gray value returned (i.e. this sample is
76% bright, that one is 40% bright). If the value is below a threshold (e.g. 5%)
I take another sample.
The code is all in my Camera35mm macros I posted a year or so ago.
Cheers,
Edouard.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Tek" <tek### [at] evilsuperbraincom> wrote:
> BTW is there any chance you could show us the source to those materials?
>
> I'd love to know if they're incredibly precisely based on reality or if they
> just look real because of the proximity pattern and how you've lit &
> "photographed" them.
I was planning on posting them in a test scene with an update to my proximity
pattern code. They are all reasonably simple. The bronze and stone ones were
already posted in 0.95 of my proximity macros I think.
As for the realism, my personal feeling is:
The lighting - it's based on a light probe, so is as real as a photograph. The
Light Dome, plus low quality, single bounce radiosity gives as good a result as
very high quality radiosity.
The textures - it's mainly the proximity pattern with some randomness (those
ones are just multi-scale bozo). The proximity pattern triggers your brains
shape recognition circuits, and the randomness makes sure the pattern doesn't
look too artificial. Making everything have even a tiny bit of reflection (even
if it is blurred to the point of "phong highlights") makes everything realistic
too.
The model - again, your brain is busy working out the complex shape, so you
don't really notice anything else as much in the scene. A viewers brain only has
finite cycles, so if you use them up they won't notice the rest of the scene's
imperfections.
The camera - probably doesn't make much difference, but making it based on a
35mm film camera and lens just ensures the field of view and depth of field
match a users expectations (based on all the real photography they've viewed).
Cheers,
Edouard.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 01/05/2010 2:31 AM, Edouard wrote:
> I was planning on posting them in a test scene with an update to my proximity
> pattern code. They are all reasonably simple. The bronze and stone ones were
> already posted in 0.95 of my proximity macros I think.
>
and would be interested in seeing any updates. Unfortunately RL has
--
Best Regards,
Stephen
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Edouard" <pov### [at] edouardinfo> wrote in message
news:web.4bdb84708fe449e421619a220@news.povray.org...
>
> The model - again, your brain is busy working out the complex shape, so
> you
> don't really notice anything else as much in the scene. A viewers brain
> only has
> finite cycles, so if you use them up they won't notice the rest of the
> scene's
> imperfections.
Actually I'm not so sure about that. Your brain's visual systems are
massively parallelized, but only the centre of your vision can see detail,
so your eyes get instructed to scan all of the details your visual system
wants more info about. Certainly your background will get less scrutiny than
the more detailed models (to paraphrase your statement: your eyes only have
finite time), but I was conciously aware of staring at the words and lines
on your background because they were visually interesting. Nothing about the
background undermined the impression that it was a photo.
Though the shape of the blur on the lines at the back did draw my attention,
it didn't *feel* wrong but conciously I know a camera aperture should give a
more even blur. It looks almost like motion blur from someone nudging the
camera when the shot was taken, though presumably it's just revealing your
sampling pattern for the focal blur. It didn't trigger my "CG" instincts,
but it's one of the first things that drew my eye.
--
Tek
http://evilsuperbrain.com
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
