Jeremy Hu’s new Autoremesher tool has been released. Jeremy is the developer of Dust3D, an accessible modeling tool that recently received an Epic Megagrant.
Autoremesher is based on open source auto-remeshers like Quadriflow, but includes some optimizations.
Hopefully the Blender developers will keep an eye on this project. As it’s open source, it might be a very interesting upgrade of Blender’s Quadriflow.
The quality of the mesh looks quite nice
Just did a test with the Fertility model with the Quadriflow implementation of Blender, some parts like the arm gets better edge flow in Quadroflow than with the Autoremesher, while Autoremesher produces better results for the rest of the body
@14AUDDIN
Its interesting to see.
I tried to subdivide a quad mesh, with edge loops which do not conform with curvature of the surface.
It does create wave-like artifacts. They could be minimized if we chose higher output mesh density.
But there are cases I rather like to avoid excessive loops. Just enough to describe forms and deformations. For this purpose topology cant be arbitrary. Edge flow needs to conform with object curvature.
By the way, this someone tried that? Because afaik the combo of tools he used should be as slow as hell… but maybe he did some optimization… gonna try it when I’m back from holidays…
EDIT: Actually I see someone did. @14AUDDIN, can you share something about times?
The screenshot I posted was from the Twitter account of the developer, not mine.
The developer said “slower than Instant Meshes but with better global edge flow, and without the Quadriflow like artifacts”.
Here’s the unstable release if you want to try it foryourself: https://github.com/huxingyi/autoremesher/releases/tag/unstable
Thanks for your Autoremesher impressions, @14AUDDIN and @Leonard_Siebeneicher. I’ve informed the developer of your posts in this thread. I guess a possible auto-retopo improvement to Blender’s Quadriflow is enough on-topic for this thread, or not?
Did some testing, this model (183k faces) took 43 seconds to remesh
Thanks for the new test. Interesting. The five-edge singularities in the Autoremesher algorithm take care of a better edge flow in branching parts, as is visible in the snout, particularly on the cheeks. This approach is comparable to Quad Remesher / ZRemesher.
Autoremesher is less successful than Quadriflow in the pointy outer part of the ears though. A closed edge loop around the eye sockets would be nice too.
Could be done if there was a way to add, remove and change the position of singularities, this is what the developer said when I asked if the user could use guides: “Not sure yet, I am still testing the balance of the constraints. The miq algorithm is very picky.” Maybe in the future, for now seems to be still in development so better wait to see what the developer changes in the future.
Here is the results for Hard Surface models:
Original geometry (19.4k faces):
I am pretty clueless about this sort of stuff… could it be possible in the future to have multiple automesher algorithms all controlled by sculpt modes face sets? Like Face sets in sculpt mode might be useful at choosing what should be automeshed? If people have trouble with one automesh algorithm that messes up things on their object - more controll might be nice
I think its nice other options are being developed anyway 
I doubt it unless you were to separate the meshes into different sections and then remesh them separately, but after that you probaly wouldn’t be able to join them since each remesher would run independently from eachother and therefore would not know how many edges should be assigned to the boundaries of each mesh.
yeah your right. 
Oh well… I think face sets are awesome maybe not for this use though
This is an option in Zremesher, you can remesh by poygroups(same concept as face sets)
Edit: sorry, didn’t realise you were asking about multiple algorithms working simultaneously on different areas of a mesh. Not possible. 
All these autoremeshing tests on organic meshes are meangingless to me. They all spit out arbitrary polysoup with spirals and random poles. Fine for sculpting, but not much else.
For me, the holy grail of autoretop is sensible, optimised hard surface production topology, rather than arbitrary dense sculpting meshes. Fast, and on the GPU(or multi-threaded)
Even the new 3dsMax autoretop solution is nowhere close to this holy grail hardsurface solution.
The spiraling and edge flow deviations are indeed still a common issue among auto-retopologizers, even in Quad Remesher / ZRemesher. But Autoremesher is only a very fresh alpha at the moment, so let’s keep an eye on it.
I’m also still lighting a candle on my auto-retopo gods altar, asking for AI / ML implementation. I think feeding lots of perfectly manually retopologized ‘ground truth’ model variations to an ML database could be the way to the holy auto-retopo grail.
Does Autoremesh work with non-manifold geometry?
Nope, doesn’t work with open meshes
Does it remesh as soon as it’s loaded. I can’t get nothing to happen ,but a circle spinning in upper corner after loaded. I tried on just a simple sphere.
