🌟🌟[AD] Try our collision-aware ACD method! (MIT license, commercial use allowed)🌟🌟

[Colin97]

We are glad to release our collision-aware ACD method that features:
(a) A novel concavity metric to detect approximation errors that introduce additional collisions in downstream applications (e.g., physics simulator). Volume-based concavity metric may be less sensitive to those approximation errors with a small volume (e.g., filling small holes, introducing thin planes).
(b) No voxelization preprocessing, directly decomposing the input mesh. If the input mesh is already a watertight manifold, we can better preserve the fine-grained details of the input shape and avoid discretization artifacts.
(c) Multi-step planning, which leads to better global solutions than the one-step greedy strategy.

[jratcliff63367]

Ok once I cloned and included the 3rd party libraries it builds.

The library appears to produce reasonable results, but it is quite slow, much slower than version 4 of V-HACD and unlike V-HACD you cannot specify the target number of output convex hulls.

With V-HACD you simply specify how many hulls you want in the output without the need for changing a tuning parameter, re-running it, then going again and again.

Nevertheless, congratulations on the project and paper.

[jratcliff63367]

What is the licensing on your implementation? It does appear to produce slightly better results than V-HACD, however, it's like 50 times slower so generally impractical to use. I might consider integrating it into the V-HACD code base if it has a reasonable license and I can make it go faster. I'm generally pretty good at code optimization; though getting a 50x perf boost would be quite a challenge.

[Glavin001]

[CoACD] does appear to produce slightly better results than V-HACD, however, it's like 50 times slower so generally impractical to use. I might consider integrating it into the V-HACD code base if it has a reasonable license and I can make it go faster.

@Colin97 @SarahWeiii (CoACD authors), @jratcliff63367 : Is there a chance you will collaborate together on a higher-quality and faster version of approximate convex decomposition? 🎉 😃

The license appears to be MIT: https://github.com/SarahWeiii/CoACD/blob/main/LICENSE

P.S. I'm working on bringing approximate convex decomposition to the web (see #105). Would love to help grow adoption of your work by empowering anyone with a web browser to use it without any additional installation.

[Colin97]

What is the licensing on your implementation? It does appear to produce slightly better results than V-HACD, however, it's like 50 times slower so generally impractical to use. I might consider integrating it into the V-HACD code base if it has a reasonable license and I can make it go faster. I'm generally pretty good at code optimization; though getting a 50x perf boost would be quite a challenge.

Yes, our license is MIT. We would be glad if you could help to optimize the code. But we would like to have a verification check before integrating it into the V-HACD code base to ensure it works as expected.

[Colin97]

[News] We have replaced the original non-commercial dependencies. So all of our code is under MIT license, and all dependencies allow commercial use now!

[jratcliff63367]

Nice! I will have to check it out.