Post Process Vs Baked Animations for Performance

Hi,

We are trying to make as much performance gains as possible and currently looking at our animation sequences.

I was curious about the relate performance between for example: Having twist joints baked down vs adding a prost-process.

My inclination is an ABP is better since the baked down tracks are duplicated across many animation sequences compared to the single ABP.

Are there any potential drawbacks to using a post-process that might make using baked down tracks preferable?

Are there even better solutions to this issue? Just want to be sure it’s worth it before starting to remove animation tracks

Thanks,

-Robert

Hey there,

The answer here is really based on your needs. At the end of the day, a couple of tracks to control twist bones and various ancillary movement will increase your animation size, but potentially (most likely) lower performance overhead on anim bps depending on how you’re handling your post-process and how many characters you have on the screen. But that comes with visual trade-offs that may matter.

Part of the reason we have post-process blueprints is for handling visual issues that come from things like modifying the pose when you use IK. If you use IK to move your hand to the forend of a rifle, you will want your twist bones to work accordingly so that the wrist and arms look good when holding the weapon. That is one example, but there are numerous others, often to do with visual fixup of the pose. Where you see the cost trade-off is also dependent on the system you use to control those visual fixups. A control rig is a great tool for these things, but it may have a higher overall cost than you’re willing to pay compared to an anim node or baking it into the animation. Because of that, I always recommend doing a performance trace to see where your performance is going, and trying out different approaches to find the right balance.

Dustin

Hey Dustin,

Thanks for the reply. To give more context, we are just currently playing with Apply a Percentage of Rotation nodes with has the fast path on it so it must be pretty performant there.

Control Rig is a bit overkill for our needs and like you mentioned has some overhead attached. We are already using ABPs for other simulation related things and so my inclination is towards the ABP while keeping the animation sequences “clean” of any joint tweaking with the added benefit that those can always be adjusted later.

So I can’t imagine those nodes being more expensive than baking down joints which can add up to 9 tracks, one for each transform axis.

I guess a question I would have where is if the Apply a Percentage of Rotation node is similar in storing all the transform axis?

Robert

I guess a question I would have where is if the Apply a Percentage of Rotation node is similar in storing all the transform axis?

Generally speaking, the cost of entering a node and doing a calculation in animbp is usually more expensive than decompressing tracks when the animation plays. In terms of comparison, it really depends on what nodes we’re talking about, as decompression is very fast and some anim nodes are really fast. That is where I say measure and validate what you’re willing to pay for in the animbp. That said, the cost here is most likely negligible and worth the visual upsides you’ll have.

Dustin

Ok makes sense. Thanks for the explanation.