On 03/11/2016 07:13 PM, William F Pokorny wrote:
>
> It looks like the difference has something to do with this test in
> <install>/source/core/render/tracepixel.cpp :
>
> //          if (cameraDirection.IsNearNull(EPSILON))
> //              return false;
>
> because if commented as shown above, I get the same result seen from the
> posted windows exe result. I'm still playing and I'll post more if I
> learn more.
>
> Bill P.

Using fabs() instead of abs() in the following two places in 
<install>/source/coure/math/vector.h fixes the Ubuntu 14.04 issue for me.

... for 2D vectors
          inline bool IsNearNull(T epsilon) const
         {
             return (fabs(vect[X]) < epsilon) &&
                    (fabs(vect[Y]) < epsilon);
         }
... for 3D vectors
         inline bool IsNearNull(T epsilon) const
         {
             return (fabs(vect[X]) < epsilon) &&
                    (fabs(vect[Y]) < epsilon) &&
                    (fabs(vect[Z]) < epsilon);
         }


I believe fabs() is the safe function to use. Must be the MS compiler in 
use did the right thing with just abs() on seeing the input is a float.

Bill P.

Post a reply to this message

>
> Using fabs() instead of abs() in the following two places in
> <install>/source/coure/math/vector.h fixes the Ubuntu 14.04 issue for
> me.
>

   Yep... that fixes it also for me. Good job, Bill... thanks!

   I will let rendering the cave in ODS format this night, then will make
the image available so those with VR glasses can test it (I did buy
yesterday some cheap Cardboard glasses for 3 euros, and lost an hour
mounting them just to discover that my Android phone is not
compatible... :).

--
jaime

Post a reply to this message

I'm happy you have addressed the Unix issue! Great!

In the mean-time, i write a draft of a Wiki page and write on the contact form
about how i can create it on the wiki.
http://www.clodo.it/host/images/53820b79ebb17bf6c2855e82e6ff97471a166b30.png

I also create a GitHub fork with the implementation in C
https://github.com/Clodo76/povray
Simply, in the current 'spherical' camera, i added two parameter: 'ipd' and
'ods'.
IPD, default 0.065.
ODS=0, The default spherical camera is used (no-stereo).
ODS=1, Left eye render (no-stereo)
ODS=2, Right eye render (no-stereo)
ODS=3, Side by Side
ODS=4, Top/Bottom

The C implementation is a little faster: i a sample scene, 19 seconds the C
implementation VS 22 seconds of the user-defined camera.

I'm thinking if i can cite my GitHub fork in the Wiki page, or it's not allowed.

I'm still looking how to implement support for the camera direction.
Current implementation use a forced <0,0,1> as direction as front view.
But with 'plants_demo_pano' rendering, i need another direction, otherwise the
cool stuff (tree) are behind me.

I hope i can post soon some nice render.

Jaime, sorry for your Cardboard... of course i can test your image with my
Oculus Rift.

Post a reply to this message

> I'm still looking how to implement support for the camera direction.
> Current implementation use a forced <0,0,1> as direction as front
> view.

  Don't know still how much sense it makes to have an arbitrary direction
vector... maybe better would a be a way to just rotate the camera around y.

> But with 'plants_demo_pano' rendering, i need another direction,
> otherwise the cool stuff (tree) are behind me.

   For the moment, you can do with adding a - sign to the z function of
the direction to reverse it: it works with my cave to have the interior
as the initial view.

--
jaime

Post a reply to this message

Jaime Vives Piqueres <jai### [at] ignoranciaorg> wrote:
>   Don't know still how much sense it makes to have an arbitrary direction
> vector... maybe better would a be a way to just rotate the camera around y.

Yes, but i aim allow people to open it's source and simply render with
spherical/ods.
plants, for example, have a
# look_at posTree+y*yCam

In my C edition, i'm thinking to take the current direction, compute the
rotation around Y axis, and apply it, without any additional parameter.

In my user_defined edition, i can add a "#declare rotationY = 0" and use it in
formulas.
The fact that user_defined have 6 different function, and not 2 that can return
a vector, it's a pain.. also force me to expand theta/phi in each function, i
cannot find a syntax to declare outside. Maybe this are the reason of the little
performance drop VS the C edition.

>    For the moment, you can do with adding a - sign to the z function of
> the direction to reverse it: it works with my cave to have the interior
> as the initial view.

Thanks. Anyway i know that: original ODS Google algorithm is right-handed, i
invert the Z to the left-handed POV-Ray coordinate system. I will comment this
for my rendering of plants.

Post a reply to this message

Am 12.03.2016 um 12:01 schrieb Clodo:

> The C implementation is a little faster: i a sample scene, 19 seconds the C
> implementation VS 22 seconds of the user-defined camera.

With more complex scenes, the relative difference between the C and SDL
implementations will probably be far smaller.

Post a reply to this message

Am 12.03.2016 um 13:22 schrieb Clodo:

> The fact that user_defined have 6 different function, and not 2 that can return
> a vector, it's a pain..

I totally agree.

Unfortunately POV-Ray's function engine can only handle scalar functions.

Post a reply to this message

> Jaime, sorry for your Cardboard... of course i can test your image
> with my Oculus Rift.

   In case you missed it, I posted a test on p.b.images... enjoy! I will
try it next weekend with the phone of my brother, just to see how it
feels to be inside my own creation.

--
jaime

Post a reply to this message

I posted an ODS video on https://www.youtube.com/watch?v=UEFO5mQ2zeg
and ODS YouTube HOWTO at https://github.com/koppi/pov-ods .

Happy POVing!

Post a reply to this message

> I posted an ODS video on https://www.youtube.com/watch?v=UEFO5mQ2zeg
> and ODS YouTube HOWTO at https://github.com/koppi/pov-ods .
>

   Very nice demo... bookmarked!

--
jaime

Post a reply to this message