top of page

Wearables for the Metaverse (3)

Weight Transfer


There are some items where transferring the existing weight of the avatar to the costume can be quite tedious, even when done by hand.


Speaking of weight transfer, approaches implemented in mainstream 3D applications include the nearest vertex method, nearest surface interpolation, and voxel-based heat conduction simulation. Still, we decided to explore whether a method exists that can withstand fully automatic implementation.


The item prepared for the test was a fairly advanced challenge: a highly multi-layered top with a T-shirt, collar, and double pockets, and wide pants with a gap of just 2 mm between the hems.

Source model in A pose



Nearest Face Interpolation:

The calculation is the simplest and most intuitive. It is best suited for clothes tailored to the body's shape. It also works with multi-layer meshes. On the other hand, the transfer of concave parts is weak, and once a vertex is bound to the wrong bone, the mesh will be stretched when animating the character, which will look like a noticeable bug.

If the hems stick together it will look buggy.


Voxel Heat Diffusion:

Simulation is performed using a voxelized clothing model based on the heat conduction principle. Since weights do not transfer to non-adjacent parts, it is expected that the limbs and body will produce better results than nearest surface interpolation.


As a result of actually doing it, the weight transfer between the joints became smoother, but it was difficult to control that range, so the effect of continuing the weight from the body seemed weak.


It's fine for tasks where neither the avatar nor the outfit is rigged, but transferring an already rigged body to an outfit seems impractical.


Tailoring has changed from Western clothing to hakama


Corrective Smooth Baker:

As a post-processing step, we used nearest surface interpolation and Corrective Smooth to detect overly deformed areas and resolve the inappropriate assignments with group weights. As a result, we softened the areas that were stuck together, but as a side effect, the consistency of weights in multi-layered fabrics was destroyed.



Original Method


Adjacent Over-deformed Refinement:

This process recursively detects over-deformed parts and overwrites the weights from the adjacent meshes. This completely solved the problem of the legs sticking together.


However, by prioritizing continuity, the chest area of female characters is recognized as separate like both legs, and the consistency is broken. Also, surface-attached objects such as pockets may fly off the surface of the clothing.


The consistency of the double breasted is teetering.


Bi-directional Segment Projection:

Even though the bottom of the pants is so close to each other, weight transfer is achieved without any sticking. The flexibility of the segments and joints is also more faithful. The execution time is short, but the continuity of the attached objects is also robust.


With this implementation, costumes can be made into assets in the metaverse through a fully automated pipeline from the CAD app, saving the industry over 70 hours of manual work per item.

Applying Motion by Mixamo
DMKTZ designs were created as UGC


For more digital fashion, visit the official DMKTZ Instagram


Piece!

Yorumlar


bottom of page