Angle based UV-unwrap: distorted UV island with even angles on geometry #78101

System Information
Operating system: Windows-7-6.1.7601-SP1 64 Bits
Graphics card: GeForce GT 640/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 445.87

Blender Version
Broken: version: 2.90.0 Alpha, branch: master, commit date: 2020-06-19 23:28, hash: 874af5301c

Short description of error

Exact steps for others to reproduce the error
2020-06-21_18-02-13.mp4
untitled.blend

Added subscriber: @Vyach

#102833 was marked as duplicate of this issue

Added subscriber: @mano-wii

Changed status from 'Needs Triage' to: 'Confirmed'

Perhaps it is a known problem since we have: #71959 (Improve UV unwrapping balance between angle and area preservation)
But I'm confirming for now.

@mano-wii
okay

Added subscriber: @no4me

Added subscriber: @iss

Adding back modelling because this is otherwise untagged with project in https://developer.blender.org/maniphest/query/J_RaaSzk5wr0/

poke

Added subscriber: @1D_Inc

I am not sure that angle-based term stands for "preserving all the angles".
I think it works different.

In #78101#1212397, @1D_Inc wrote:
I am not sure that angle-based term stands for "preserving all the angles".
I think it works different.

There shouldn`t be different behavior for similar angles, because it become unpredictable (what angle will be distorted?)

In #78101#1212518, @Vyach wrote:

There shouldn`t be different behavior for similar angles, because it become unpredictable (what angle will be distorted?)

Angle based method is supposed to deal with organic/irregular topology, while conformal - with linear. Maybe it is a naming problem.

Added subscriber: @Chris_Blackbourn

I actually spotted this one in the code first, so it's nice to have a real world example.

I think what's happening here is those quads are being triangulated before being passed to the solver. By triangulating, we create additional constraints across the diagonals not present in the original mesh.

I've constructed a fully symmetric model which is specially designed to demonstrate the problem here:

Note that a fully symmetric unwrap can be created by hand, keeping that square a square, and something like this would be the desired unwrapping:

Blender file for the curious. You can try scaling in the 'Y' for some interesting results, splitting the quad, or removing some of the triangles. UnlikelyShape.blend

In terms of fixes, if this is the cause of the bug, we would need to upgrade the LSCM solver to operate directly on quads as well as triangles.

I just skimmed through the original "Least Squares Conformal Maps
for Automatic Texture Atlas Generation" paper and it doesn't mention quads, so this would be either a literature search or some original research or perhaps a "contact-the-author" situation.

@Chris_Blackbourn interesting case.
In 2.7 angle based method give a better result with such kind of a geometry (at least tries to), while in 3.0 both methods give the same results.

Turns out to be more subtle than anticipated. Please contact me if you're working on this one.

See another instance of the bug here: https://developer.blender.org/T102833

Added subscriber: @070NRG

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Angle based UV-unwrap: distorted UV island with even angles on geometry #78101