Animation delta (change) render only??
Posted: Wed Oct 28, 2015 2:40 am
Hi Otoy team,
This may be something already on the To-Do list or may be a low-tech compositing solution.
I had a sunrise idea this morning regarding saving some effort rendering animations: the theory went like this.
Could the Delta/change within an animated scene/sequence be used to rationalised work for each frame based on how accurate the previous frame bounce calculations would be for the current frame. Anything below a certain 'accuracy' threshold discarded and re-rendered. obviously this isn't 'blind' or 'purists' unbiased rendering; but I thought it could save 10-30% on each frame... right?
eg1. largely static scene with consistant camera pan in one direction: frame 2 is likely to be very high percentage accurate to frame 1... seems like a lot of calculations to throw away to achieve such a small shift... obviously frame 20 is much less similar to frame 1: which is why the threshold is important.
eg2. Largely static scene with single moving object; lets say for argument that the object occupies 5% of the visible FOV and is about 2/3 deep into the field. I'd like to argue that the perceptible influence of this object is less than 50-60%.. but this isn't my field of expertise. As above: understand I'm talking about biasing the result..
eg3.The compositing idea: and I'm almost certain this is possible either now or in v3.0. could the static background be rendered to a big HDRI then just the moving object rendered on an alpha with the static background HDRi used to get good reflections/light on the object itself?
eg.4 is I suppose a combination of the above. or perhaps the threshold for re-sample looks at a per-object control to define if bounces that touch the object are weighted towards resampling or reusing based on the preference of the artist....
This may be something already on the To-Do list or may be a low-tech compositing solution.
I had a sunrise idea this morning regarding saving some effort rendering animations: the theory went like this.
Could the Delta/change within an animated scene/sequence be used to rationalised work for each frame based on how accurate the previous frame bounce calculations would be for the current frame. Anything below a certain 'accuracy' threshold discarded and re-rendered. obviously this isn't 'blind' or 'purists' unbiased rendering; but I thought it could save 10-30% on each frame... right?
eg1. largely static scene with consistant camera pan in one direction: frame 2 is likely to be very high percentage accurate to frame 1... seems like a lot of calculations to throw away to achieve such a small shift... obviously frame 20 is much less similar to frame 1: which is why the threshold is important.
eg2. Largely static scene with single moving object; lets say for argument that the object occupies 5% of the visible FOV and is about 2/3 deep into the field. I'd like to argue that the perceptible influence of this object is less than 50-60%.. but this isn't my field of expertise. As above: understand I'm talking about biasing the result..
eg3.The compositing idea: and I'm almost certain this is possible either now or in v3.0. could the static background be rendered to a big HDRI then just the moving object rendered on an alpha with the static background HDRi used to get good reflections/light on the object itself?
eg.4 is I suppose a combination of the above. or perhaps the threshold for re-sample looks at a per-object control to define if bounces that touch the object are weighted towards resampling or reusing based on the preference of the artist....