 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/06/2012 03:44 PM, clipka wrote:
> Am 06.12.2012 19:48, schrieb James Holsenback:
>> On 12/06/2012 01:15 PM, Le_Forgeron wrote:
>>> Le 06/12/2012 17:39, James Holsenback nous fit lire :
>>>>
>>>> play around with the scaling and see when it starts to call it finite
>>>> ;-)
>>>
>>> I don't play... I cheat & I looked at the code.
>>> In parse.cpp, BOUNDS_VOLUME is an inline function from bbox.h which
>>> compute the volume of the bounding box (simple product of height, width
>>> & depth).
>>> INFINITE_VOLUME is the limit to set the infinite flag.
>>> The value of INFINITE_VOLUME is BOUND_HUGE. (parse.cpp)
>>> The value of BOUND_HUGE is 2.0e+10 (configbackend.h)
>>>
>>> So, the sphere is to become finite when the box (true cube) is to have a
>>> volume of 2.0e+10 or less. That's for a side of 2714.4176 (cubic root of
>>> 2.0e+10), so for a radius of 1357.2088
>>
>> Yes ... obviously the program is catching something, I guess I'm
>> challenging /what/ it's catching. Again I'm encouraging you to /play/
>> around with the scaling in my example and note /when/ it starts calling
>> the object finite ;-)
>
> So, what are the results of /your/ playing around that makes you think
> we should bother to try it as well?
>
LOL ... starting with the simple scene it does not report finite until
scale is set to 1000
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Am 06.12.2012 22:16, schrieb James Holsenback:
> LOL ... starting with the simple scene it does not report finite until
> scale is set to 1000
From the math, the threshold should be between 1357 and 1358 if the
sphere is just transformed and/or rotated. (If a rotation is applied via
an arbitrary matrix, the threshold might be as low as 783 to 784 in
extreme cases.)
An infinite object is not excluded from bounding /per se/, but does not
benefit from /hierarchical/ bounding.
(Fun fact: Spheres have such a simple math that they skip bounding box
testing entirely, so they only benefit from hierarchical bounding at all.)
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/07/2012 12:20 AM, clipka wrote:
> Am 06.12.2012 22:16, schrieb James Holsenback:
>
>> LOL ... starting with the simple scene it does not report finite until
>> scale is set to 1000
>
> From the math, the threshold should be between 1357 and 1358 if the
> sphere is just transformed and/or rotated. (If a rotation is applied via
> an arbitrary matrix, the threshold might be as low as 783 to 784 in
> extreme cases.)
>
> An infinite object is not excluded from bounding /per se/, but does not
> benefit from /hierarchical/ bounding.
>
> (Fun fact: Spheres have such a simple math that they skip bounding box
> testing entirely, so they only benefit from hierarchical bounding at all.)
>
Seems that everyone is OK with this ... I don't recall this limitation
in the past. Could this be side-effect of a recent change?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Am 07.12.2012 12:34, schrieb James Holsenback:
> On 12/07/2012 12:20 AM, clipka wrote:
>> Am 06.12.2012 22:16, schrieb James Holsenback:
>>
>>> LOL ... starting with the simple scene it does not report finite until
>>> scale is set to 1000
>
> Seems that everyone is OK with this ... I don't recall this limitation
> in the past. Could this be side-effect of a recent change?
Nothing truly recent, that's for sure. Did you test 3.6 behaviour?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 7-12-2012 12:34, James Holsenback wrote:
> Seems that everyone is OK with this ... I don't recall this limitation
> in the past. Could this be side-effect of a recent change?
I feel a bit concerned though. I did not test your code, but what does
it mean for the render?
Thomas
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/07/2012 08:33 AM, Thomas de Groot wrote:
> On 7-12-2012 12:34, James Holsenback wrote:
>> Seems that everyone is OK with this ... I don't recall this limitation
>> in the past. Could this be side-effect of a recent change?
>
> I feel a bit concerned though. I did not test your code, but what does
> it mean for the render?
>
> Thomas
>
Well this all started when I noticed "1" infinite object reported when I
knew that I had none, or at least I thought ... that led to the
discovery that my skydome object (generally just has a hdr image) was
the culprit. In the past I usually had scale in the 100 to 500 range but
recently I found the need for 1000+ ...
The guys who would know more about such things don't seem alarmed, and
I'm quite content calling my skydome object infinite when it's that big
... new (lol or new to me) paradigm right!
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 7-12-2012 15:25, James Holsenback wrote:
> Well this all started when I noticed "1" infinite object reported when I
> knew that I had none, or at least I thought ... that led to the
> discovery that my skydome object (generally just has a hdr image) was
> the culprit. In the past I usually had scale in the 100 to 500 range but
> recently I found the need for 1000+ ...
>
> The guys who would know more about such things don't seem alarmed, and
> I'm quite content calling my skydome object infinite when it's that big
> ... new (lol or new to me) paradigm right!
I had not noticed the /infinity/ of some of my objects before and they
seem to render normally indeed. In /The Sultana/ scene there are the sea
and the Earth below for instance, as well as the atmosphere which are on
the 10e5 radius level.
Thomas
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Thomas de Groot wrote:
> On 7-12-2012 12:34, James Holsenback wrote:
>> Seems that everyone is OK with this ... I don't recall this limitation
>> in the past. Could this be side-effect of a recent change?
>
> I feel a bit concerned though. I did not test your code, but what does
> it mean for the render?
The limit seems a bit arbitrary, considering that it depends on the
scene scale. I suspect this means that in some cases a scene can render
slower if it is modelled, e.g., in mm per unit vs. m per unit (in the
former case an object in the 1000 - 10000 range may not be uncommon).
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
clipka <ano### [at] anonymous org> wrote:
> Am 06.12.2012 19:30, schrieb Warp:
> > Le_Forgeron <jgr### [at] freefr> wrote:
> >> an object is considered infinite when it is big enough...
> >
> > What exactly is the advantage of this?
> If an object is virtually everywhere, testing against its bounding box
> is wasted computing time.
But the volume of the object doesn't tell this. The camera could well be
located such that the object in question is very small on screen. And
conversely, even a small object could fill up the entire view (eg. because
the camera is inside or extremely close to it.)
Shouldn't the algorithm be slightly more involved? In other words, try
to estimate the size of the object with respect to the camera position
and direction, rather than simply as a function of its volume.
--
- Warp
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Le 2012-12-07 07:39, clipka a écrit :
> Am 07.12.2012 12:34, schrieb James Holsenback:
>> On 12/07/2012 12:20 AM, clipka wrote:
>>> Am 06.12.2012 22:16, schrieb James Holsenback:
>>>
>>>> LOL ... starting with the simple scene it does not report finite until
>>>> scale is set to 1000
>>
>> Seems that everyone is OK with this ... I don't recall this limitation
>> in the past. Could this be side-effect of a recent change?
>
> Nothing truly recent, that's for sure. Did you test 3.6 behaviour?
>
I remember some renders using 3.6 where there was no infinits objecte
that reported some infinite objects. Those where mostly large "world
spheres" or other objects that would show wherever the background would
have been visible if those where not present.
My first impression was that when an object bounding box or extent
totaly enclosed the camera, then that object would be treated as infinite.
Alain
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
