Page MenuHome

Cycles - Inconsistent Motion blur based on interpolation mode
Closed, DuplicatePublic

Description

System Information
Operating system: Linux-5.3.0-29-generic-x86_64-with-debian-buster-sid 64 Bits
Graphics card: GeForce GTX 1080 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 440.59

Blender Version
Broken: version: 2.82 (sub 6), branch: master, commit date: 2020-02-11 14:45, hash: rBc939b4df18e9
Worked: (optional)

Short description of error
The 'look' of motion blur changes (e.g. frame 1 and frame 2) based on the interpolation mode used, despite there being no room to interpolate between the frames.

Above are frames 1 and 2 of a test animation. The cube is parented to the 'child' bone (using the armature modifier). The 'child' bone is parented to the 'parent' bone as an armature relationship.

All of the bones of both armatures have exactly the same keyframe values, but the interpolation mode of the child bone is different. In the top rig, the child bone has 'constant' interpolation. In the bottom example, the child bone has 'bezier' interpolation.

I can understand why the interpolation mode will have an influence on the look of motion blur of keyframes not directly next to each other, but neighbouring keyframes like this should give exactly the same result as there's no room for interpolation.

Exact steps for others to reproduce the error

  • Open 'blur-test.blend'
  • Hit render

Event Timeline

My understanding is that the keyframe interpolation of the parent bone is being taking into account rather than just that of the child bone. This leads to issues with motion blur when rendering rigged characters with varying keyframe interpolations on different bones. We have noted extra long render times, and odd looking motion blur when this happens.

This is ultimately same issue as T73756.

I can understand why the interpolation mode will have an influence on the look of motion blur of keyframes not directly next to each other, but neighbouring keyframes like this should give exactly the same result as there's no room for interpolation.

I think, that motion blur will sample position of object at some (arbitrary) resolution, and calculate speed for each sample. Bottom cube has to travel greater distance between 2 samples so blurred trail is weaker. I haven't looked at motion blur design or code to be 100% sure here, but I bet this is pretty good description based on my observations.

The cubes are travelling the same distance. This is the whole reason the the concern has been raised

If you have a parent bone with bezier interpolation between sequential keyframes moving a child bone with bezier interpolation between sequential keyframes, you get one result.
If you have a parent bone with constant interpolation between sequential keyframes moving a child bone with bezier interpolation between sequential keyframes, you get a different and impossible result.
This holds even if the visual distance travelled by the child bone is exactly the same in each test. There is no good reason for the interpolation of the parent bone to be taken into account in any blur calculations when the child is key-framed and animated and has its own keyframe interpolation.

If you have a parent bone with bezier interpolation between sequential keyframes moving a child bone with bezier interpolation between sequential keyframes, you get one result.

You have to account for motion while camera shutter is open.
This does apply not only to bones, but any means to teleport object. If you do so, while camera shutter is open, it wil be interpreted as motion with high speed, because motion is sampled.

If you have a parent bone with constant interpolation between sequential keyframes moving a child bone with bezier interpolation between sequential keyframes, you get a different and impossible result.

You can drive object position with non-deterministic methods, this is not just technical limitation of current design.
It is impossible result for impossible movement.

in the screen shot both child bones have the same movement in relation to the camera. Also the parent bones move the same amount in relation to the camera but the parents have different interpolation , this ends up create two different motion blur results on the object parented to the child. Multiply that effect on a rig with numerous parents and the results are all over the place.