Depth Of Field And Motion Blur

[user_deleted]

Which comes first in the composite pipleline, depth of field or motion blur? Right now I think depth of field gets calculated first, then defocuses.

[user_deleted]

interesting question...

theoretically i'd say that motion blur needs to be applied first because the objects are already streaked when the hit the lens, however, the differences seem to be pretty much zero, both in normal gamma corrected space and in linear float:

CODE
RGrad1 = RGrad(720, 576, 2, 361.25, 288, 1, 23, 3, 0.5, 1, 1,
    1, 1, 0, 0, 0, 0, 0, 0);
RGrad2 = RGrad(720, 576, 4, 361.25, 288, 1, 23, 3, 0.5, 1, 1,
    1, 1, 0, 0, 0, 0, 0, 0);
Defocus1_clone1 = Defocus(RGrad1, Defocus1.xPixels, Defocus1.yPixels,
    Defocus1.channels, Defocus1.percent, Defocus1.shape, Defocus1.boostPoint,
    Defocus1.superWhite);
Gamma3 = Gamma(RGrad2, 0.5, rGamma, rGamma, 1);
Move2D1 = Move2D(RGrad1, Hermite(0,[-339,98.69,98.69]@1,[339.936,98.69,98.69]@15),
    0, Hermite(0,[0,0,0]@1,[0,0,0]@15), 1, Hermite(0,[1,0,0]@1,[1,0,0]@15),
    Hermite(0,[1,0,0]@1,[1,0,0]@15), 0, 0, Hermite(0,[360,0,0]@1,[360,0,0]@15),
    Hermite(0,[288,0,0]@1,[288,0,0]@15), "default", xFilter,
    "trsx", 0, 5, 0.5, 0, 0, time);
Brightness3 = Brightness(Gamma3, 2);
Defocus1 = Defocus(Move2D1, 8, xPixels/GetDefaultAspect(), "rgba",
    100, "circle", 0.95, 1);
Move2D1_clone1 = Move2D(Defocus1_clone1, Move2D1.xPan, Move2D1.yPan,
    Move2D1.angle, Move2D1.aspectRatio, Move2D1.xScale, Move2D1.yScale,
    Move2D1.xShear, Move2D1.yShear, Move2D1.xCenter, Move2D1.yCenter,
    Move2D1.xFilter, Move2D1.yFilter, Move2D1.transformationOrder,
    Move2D1.inverseTransform, Move2D1.motionBlur, Move2D1.shutterTiming,
    Move2D1.shutterOffset, Move2D1.useReference, Move2D1.referenceFrame);
Defocus1_clone2 = Defocus(Brightness3, Defocus2.xPixels, Defocus2.yPixels,
     Defocus2.channels, Defocus2.percent,  Defocus2.shape, Defocus2.boostPoint,
    Defocus2.superWhite);
Move2D2 = Move2D(Brightness3, Hermite(0,[-339,98.3,98.3]@1,[333.936,98.68,98.68]@15),
    Hermite(0,[0,0,0]@15), Hermite(0,[0,0,0]@1,[0,0,0]@15), 1,
    Hermite(0,[1,0,0]@1,[1,0,0]@15), Hermite(0,[1,0,0]@1,[1,0,0]@15),
    0, 0, Hermite(0,[360,0,0]@1,[360,0,0]@15), Hermite(0,[288,0,0]@1,[288,0,0]@15),
    "default", xFilter, "trsx", 0, 5, 0.5, 0, 0, time);
Defocus2 = Defocus(Move2D2, 8, xPixels/GetDefaultAspect(), "rgba",
    100, "circle", 0.95, 1);
Move2D1_clone2 = Move2D(Defocus1_clone2, Move2D2.xPan, Move2D2.yPan,
    Move2D2.angle, Move2D2.aspectRatio, Move2D2.xScale, Move2D2.yScale,
    Move2D2.xShear, Move2D2.yShear, Move2D2.xCenter, Move2D2.yCenter,
     Move2D2.xFilter,  Move2D2.yFilter,  Move2D2.transformationOrder,
    Move2D2.inverseTransform, Move2D2.motionBlur, Move2D2.shutterTiming,
    Move2D2.shutterOffset, Move2D2.useReference, Move2D2.referenceFrame);
Brightness2_clone2 = Brightness(Move2D1_clone2, Brightness4.value);
Brightness4 = Brightness(Defocus2, 0.5);
Gamma2_clone2 = Gamma(Brightness2_clone2, Gamma4.rGamma, Gamma4.gGamma,
    Gamma4.bGamma, Gamma4.aGamma);
Gamma4 = Gamma(Brightness4, 2, rGamma, rGamma, 1);

[joojaa]

id first think say the other way around (after all even the source is to be blurred). In actuality it is allmost a inseparable compound so in reality both should be as fake

Reasoning:

the real situation is never blurred, but rather when captured by a sensor it smears on it because it takes time to expose the result on it. so the smearing happens on the sensor, offcourse the sensor is after the lense. thus the smearing is affected by dof at this point allready.

I think its a close linear combination tough, within their error from real. Basically the prblem is that you dont account very well for the change in the dof. But if you can then ist true what zem says streak first. As the other option is that you dof a vector step every so and so steps and dof each one. (but to get it better youfd need 3d vectors, because even z changes! but a normal motion vector dont necessarily have this data), ill try this as it sounds interesting and could be better! (and offer a slight from behid object seen dof!)

But im pretty open for any other interpretations.

[user_deleted]

I'm planning a 10-20 minute short, (preferably all in one take, no cuts) primarily rendering out of mental ray. I've been looking at zDepthDOF http://homepage.mac.com/bauer/portfolio\_sh.../zDepthDOF.html to do the dof in shake, and Im_2DMV http://www.alamaison.fr/3d/lm\_2DMV/lm\_2DMV\_ref.htm for the motion blur.

I was also a little confused about where normal lighting in post would fall in the order, then came to the conclusion that it would be before all DOF and motion blur.

[user_deleted]

QUOTE(Joojaa @ 12/20/07, 11:01 PM) [snapback]278517[/snapback]
id first think say the other way around (after all even the source is to be blurred). In actuality it is allmost a inseparable compound so in reality both should be as fake

Reasoning:

the real situation is never blurred, but rather when captured by a sensor it smears on it because it takes time to expose the result on it. so the smearing happens on the sensor, offcourse the sensor is after the lense. thus the smearing is affected by dof at this point allready.

hmm, yeah that sounds reasonable too...

if you take the sample i posted above and take the difference between the two appoaches they are very close to zero though, so i'd say best would be to try again with a real world example of the desired footage type and workflow and see if there's a bigger difference... and then just take the one that looks better
++ chris