BLI_math: improve the threshold for wrapping rotation in compatible_eul #104856

Campbell Barton · 2023-02-17T09:12:01+01:00

Campbell Barton commented

2023-02-17 09:12:01 +01:00

Having a threshold well above PI would result in discontinuity in some cases.

The discontinuity can be measured by generating euler values (both random and interpolated rotations), then comparing the accumulated difference.
Changing the threshold for wrapping rotations produces at least as good or more compatible results.

This was reported as #17297 and fixed in 0, however the change was only applied for the game-engine.

Ref !104856

The attached script can be run from the command line, reporting:

blender -b --python  measure_compat_eul.py

The output is:

Interpolated eulers 1.1463990949094296
Random eulers 4.888162368731574

After the change the output is:

Interpolated eulers 1.1338555704057216
Random eulers 4.614090494234115

(lower is better).

This is a simpler case:

from mathutils import Euler

e_prev = Euler((-12.991190910339355, -48.46185302734375, 41.59339904785156))
e_test = Euler((-16.96780014038086, 1.3109626770019531, -6.306068420410156))
e_compat = e_test.copy()
e_compat.make_compatible(e_prev)
print(tuple(e_compat), "difference:", sum([abs(a - b) for a, b in zip(e_prev, e_compat)]))

# Before: (-16.96780014038086, -48.95452117919922, 43.959415435791016) difference: 6.835293769836426
# After: (-10.684614181518555, -48.95452117919922, 43.959415435791016) difference: 5.165261268615723

Having a threshold well above PI would result in discontinuity in some cases. The discontinuity can be measured by generating euler values (both random and interpolated rotations), then comparing the accumulated difference. Changing the threshold for wrapping rotations produces at least as good or more compatible results. This was reported as #17297 and fixed in [0], however the change was only applied for the game-engine. Ref !104856 [0]: ab44742cf36639b73f1aa2c18aa55b14ff059acd ---- The attached script can be run from the command line, reporting: ``` blender -b --python measure_compat_eul.py ``` The output is: ``` Interpolated eulers 1.1463990949094296 Random eulers 4.888162368731574 ``` After the change the output is: ``` Interpolated eulers 1.1338555704057216 Random eulers 4.614090494234115 ``` (lower is better). This is a simpler case: ``` from mathutils import Euler e_prev = Euler((-12.991190910339355, -48.46185302734375, 41.59339904785156)) e_test = Euler((-16.96780014038086, 1.3109626770019531, -6.306068420410156)) e_compat = e_test.copy() e_compat.make_compatible(e_prev) print(tuple(e_compat), "difference:", sum([abs(a - b) for a, b in zip(e_prev, e_compat)])) # Before: (-16.96780014038086, -48.95452117919922, 43.959415435791016) difference: 6.835293769836426 # After: (-10.684614181518555, -48.95452117919922, 43.959415435791016) difference: 5.165261268615723 ```

measure_compat_eul.py

2.7 KiB

🚀 1

Campbell Barton commented

2023-02-17 09:13:12 +01:00

@blender-bot build linux

Campbell Barton force-pushed pr-euler-compat from 5ec3796d74 to 2f59e96d37

2023-02-17 09:18:17 +01:00

Compare

Campbell Barton changed title from ~~WIP BLI_math: correct the threshold for wrapping rotation in compatible_eul~~ to WIP BLI_math: improve the threshold for wrapping rotation in compatible_eul

2023-02-17 09:25:12 +01:00

Campbell Barton changed title from ~~WIP BLI_math: improve the threshold for wrapping rotation in compatible_eul~~ to BLI_math: improve the threshold for wrapping rotation in compatible_eul

2023-02-17 09:47:28 +01:00

Campbell Barton referenced this issue from a commit

2023-02-17 09:52:46 +01:00

BLI_math: correct the threshold for wrapping rotation in compatible_eul

Campbell Barton commented

2023-02-17 09:53:15 +01:00

Committed c6ea00de3c, closing.

Committed c6ea00de3cd4ee01a5c2070f5aadcf0aa7b661c4, closing.

Campbell Barton closed this pull request

2023-02-17 09:53:16 +01:00

Campbell Barton deleted branch pr-euler-compat

2023-02-17 09:53:25 +01:00

Clément Foucault commented

2023-02-21 12:26:08 +01:00

Following up on this. It seems like just doing this is even doing better.

void compatible_eul(float eul[3], const float oldrot[3])
{
  for (uint i = 0; i < 3; i++) {
    eul[i] = angle_wrap_rad(eul[i] - oldrot[i]) + oldrot[i];
  }
}

This is the result:

Interpolated eulers 1.1338555291295052
Random eulers 4.614090291559696

Comparing with higher precision (ROTATIONS = 10000)

// Simple wrap
Interpolated eulers 0.8127715403175354
Random eulers 4.698880729122162
// Current wrap
Interpolated eulers 0.812771551511623
Random eulers 4.698880809928011

Following up on this. It seems like just doing this is even doing better. ``` void compatible_eul(float eul[3], const float oldrot[3]) { for (uint i = 0; i < 3; i++) { eul[i] = angle_wrap_rad(eul[i] - oldrot[i]) + oldrot[i]; } } ``` This is the result: ``` Interpolated eulers 1.1338555291295052 Random eulers 4.614090291559696 ``` Comparing with higher precision (`ROTATIONS = 10000`) ``` // Simple wrap Interpolated eulers 0.8127715403175354 Random eulers 4.698880729122162 // Current wrap Interpolated eulers 0.812771551511623 Random eulers 4.698880809928011 ```

Pull request closed

This pull request cannot be reopened because the branch was deleted.

Sign in to join this conversation.

No reviewers

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

BLI_math: improve the threshold for wrapping rotation in compatible_eul #104856

Pull request closed