acceleration blur?

   5875   7   2
User Avatar
Member
1705 posts
Joined: 3月 2020
Offline
Hi,


I just stumbled into this awesome feature of the htoa translator, called acceleration blur –

https://support.solidangle.com/display/AFHUG/Geometry+Properties+%3A%3A+Motion+Blur [support.solidangle.com]

This allows curved motion blurs for velocity vector-based motions (which is very useful for thing like sparks, etc). I'm not aware of similar in “regular” Houdini (but maybe I'm not up to date), do you guys think this could be done somehow in mantra?


cheers,
imre
Imre Tuske
FX Supervisor | Senior FXTD @ Weta FX

qLib -- Houdini asset library
http://qlab.github.io/qLib/ [qlab.github.io]
https://www.facebook.com/qLibHoudini [www.facebook.com]
User Avatar
Member
143 posts
Joined: 2月 2012
Offline
Just saw this in Arnold too, looks nice. Perhaps doable in a cvex procedural?
User Avatar
Member
76 posts
Joined: 9月 2011
Online
There was this procedural which does something similar, however it no longer seems to be updated.
https://code.google.com/p/timeblender/ [code.google.com]


Ideally, if Mantra could take id attributes and deform based on them that would work the best.
If it finds matching ids on previous and next sample, the point can be interpolated along a quadratic curve defined by those 3 positions.
If it only finds previous or next and not both, it could still do a linear blur.

This would help a lot with large point counts as it could keep both .bgeo files light wihtout having to store v (and accel) and ifd files as well as you wouldn't have to bake multiple copies of the geo in for deformation blur.
User Avatar
Member
8766 posts
Joined: 7月 2007
Offline
cvex procedural would be the best, unfortunately we can't run CVEX on delay loaded geo yet, which is super bad, you can have some luck with hengine procedural, but that possibly takes license so maybe not
Tomas Slancik
FX Supervisor
Method Studios, NY
User Avatar
Member
678 posts
Joined: 7月 2005
Offline
jkunz07
There was this procedural which does something similar, however it no longer seems to be updated.
https://code.google.com/p/timeblender/ [code.google.com]

This is author of the project => http://forums.odforce.net/user/744-symek/ [forums.odforce.net].
User Avatar
Member
143 posts
Joined: 2月 2012
Offline
I dont think you would even need to know the last and next frame, you can just take the P and the V then add the accel or force of the particle multiplied the time increment (forward and back of the central time sample). Do that incremental for each geo time sample and you should get the same results as the Arnold accel blur.
So no need to worry about changing point count!
User Avatar
Member
76 posts
Joined: 9月 2011
Online
JordanWalsh
I dont think you would even need to know the last and next frame, you can just take the P and the V then add the accel or force of the particle multiplied the time increment (forward and back of the central time sample). Do that incremental for each geo time sample and you should get the same results as the Arnold accel blur.
So no need to worry about changing point count!

from my understanding, v and accel are vectors pointing towards the next and previous frame. To keep file storage down I was thinking that just keeping P and id attributes would be a smaller storage footprint then P, v and accel.

Mantra could then derive v and accel at rendertime just based off of the id attribute.

For most cases it might not be a huge deal, but if your rendering millions of points it definitely makes a big difference.
User Avatar
Member
8766 posts
Joined: 7月 2007
Offline
P, v and accel make more sense than P, v and id
as many times you have geometry not only particles and it's much safer to be independent of other frames, than having to rely on them

as well having to match by id is not as straightforward as it seems, you need to do at least 2 lookups to account for all cases (particles being born, particles existing in both and particles dying in next sample) which takes more memory and computing power than doing it straightforwardly from v and accel at the expense of some storage space
Tomas Slancik
FX Supervisor
Method Studios, NY
  • Quick Links