Differentiable Voxelization of Surface Representations

SIGGRAPH 2026 Conference Proceedings
Tobias Djuren · Ugo Finnendahl* · Markus Worchel* · Hendrik Meyer · Marc Alexa

*Equal contribution (shared second author)

About

This repository contains the official implementation of the paper "Differentiable Voxelization of Surface Representations". The paper introduces an efficient method for converting the surface representation of a shape, in particular a triangle mesh, into a volumetric representation in a differentiable manner. The resulting voxel grid of winding numbers makes it straightforward to define volumetric objectives that drive the optimization of the underlying surface (e.g. for intersection avoidance).

The algorithms are implement in C++, targeting the CPU (a CUDA GPU implementation is work in progress). The dvx package exposes these algorithms to Python, where they are readily usable with PyTorch.

Getting Started

The easiest way to install the Python package is via pip

pip install dvx-python

Optional: Test the Installation

To test the installation or a local development build, in the cloned repository run:

pip install numpy pytest svgpathtools trimesh
python -m pytest ./tests -v

Some tests may be skipped, depending on the availability of packages.

Usage

This is a minimal PyTorch example that demonstrates triangle mesh voxelization in $\mathbb{R}^3$:

import dvx.torch as dvx

# Triangle mesh as indexed face set within the [-1,1]^3 cube (v: vertices, f: faces)
v, f = ...
v.requires_grad_(True)

# Resolution of the voxel grid (same resolution in all dimensions)
n = 64 

# Voxelize the mesh -> returns a grid with shape (n,n,n) of smoothed winding numbers
voxels = dvx.voxelize(n, v, f)

loss = some_loss(voxels)
loss.backward() # Gradients are propagated to the mesh vertices v

For more usage examples, see the demos folder. The demos include examples for bandsaw cutting, space tilings in $\mathbb{R}^3$ and for self-intersection resolving to generate results similar to the ones in the paper.

Conventions for Coordinates and Data Layout

The volume considered for voxelization is the $[-1, 1]^d$ cube, where $d$ is the dimensionality of the space. The input shape (e.g., a triangle mesh) must be fully contained within this volume, otherwise the output is undefined $\space^1$.

The voxel grid covers this volume exactly, which means it reaches from $(-1, -1, \ldots, -1) \in \mathbb{R}^d$ to $(1, 1, \ldots, 1) \in \mathbb{R}^d$. We use standard naming conventions to denote points in space: $(x, y)$ for $\mathbb{R}^2$ and $(x, y, z)$ for $\mathbb{R}^3$.

For two-dimensional input and a resolution of $w$ in x-direction and $h$ in y-direction, the result is an array of shape $(h, w)$. Similarly, for three-dimensional input with the same resolutions in x- and y-directions, and a resolution of $d$ in z-direction, the result is an array of shape $(d, h, w)$. The voxel data is stored in row-major order, i.e., the index in x-direction is the fastest varying, followed by y and z.

The value of a voxel is the smoothed winding number field evaluated at its center.

The following sketch summarizes our conventions for coordinates and the data layout in $\mathbb{R}^2$ and $\mathbb{R}^3$:

$\space^1$ This restriction is not a limitation of the method itself, but rather a consequence of how the closed-form integration is currently implemented. It will be lifted in a future release.

License and Copyright

The code in this repository is provided under a BSD 3-clause license.

Citation

If you use this code or our method in your research, please cite our paper:

@inproceedings{Djuren:2026:DVX,
    author = {Djuren, Tobias and Finnendahl, Ugo and Worchel, Markus and Meyer, Hendrik and Alexa, Marc},
    title = {Differentiable Voxelization of Surface Representations},
    year = {2026},
    isbn = {9798400725548},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3799902.3811203},
    doi = {10.1145/3799902.3811203},
    booktitle = {Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers},
    articleno = {22},
    numpages = {12},
    keywords = {differentiable voxelization, efficient voxelization, smoothed winding numbers, shape optimization},
    location = {
    },
    series = {SIGGRAPH Conference Papers '26}
}

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

dvx_python-0.1.1.tar.gz (2.3 MB view details)

Uploaded Jul 20, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

dvx_python-0.1.1-cp314-cp314t-win_amd64.whl (108.2 kB view details)

Uploaded Jul 20, 2026 CPython 3.14tWindows x86-64

dvx_python-0.1.1-cp314-cp314t-musllinux_1_2_x86_64.whl (217.5 kB view details)

Uploaded Jul 20, 2026 CPython 3.14tmusllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (119.7 kB view details)

Uploaded Jul 20, 2026 CPython 3.14tmanylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp314-cp314t-macosx_11_0_arm64.whl (91.5 kB view details)

Uploaded Jul 20, 2026 CPython 3.14tmacOS 11.0+ ARM64

dvx_python-0.1.1-cp314-cp314t-macosx_10_14_x86_64.whl (105.6 kB view details)

Uploaded Jul 20, 2026 CPython 3.14tmacOS 10.14+ x86-64

dvx_python-0.1.1-cp314-cp314-win_amd64.whl (104.2 kB view details)

Uploaded Jul 20, 2026 CPython 3.14Windows x86-64

dvx_python-0.1.1-cp314-cp314-musllinux_1_2_x86_64.whl (210.3 kB view details)

Uploaded Jul 20, 2026 CPython 3.14musllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (112.7 kB view details)

Uploaded Jul 20, 2026 CPython 3.14manylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp314-cp314-macosx_11_0_arm64.whl (88.0 kB view details)

Uploaded Jul 20, 2026 CPython 3.14macOS 11.0+ ARM64

dvx_python-0.1.1-cp314-cp314-macosx_10_14_x86_64.whl (102.1 kB view details)

Uploaded Jul 20, 2026 CPython 3.14macOS 10.14+ x86-64

dvx_python-0.1.1-cp313-cp313-win_amd64.whl (101.0 kB view details)

Uploaded Jul 20, 2026 CPython 3.13Windows x86-64

dvx_python-0.1.1-cp313-cp313-musllinux_1_2_x86_64.whl (210.3 kB view details)

Uploaded Jul 20, 2026 CPython 3.13musllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (112.7 kB view details)

Uploaded Jul 20, 2026 CPython 3.13manylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp313-cp313-macosx_11_0_arm64.whl (88.0 kB view details)

Uploaded Jul 20, 2026 CPython 3.13macOS 11.0+ ARM64

dvx_python-0.1.1-cp313-cp313-macosx_10_14_x86_64.whl (102.1 kB view details)

Uploaded Jul 20, 2026 CPython 3.13macOS 10.14+ x86-64

dvx_python-0.1.1-cp312-cp312-win_amd64.whl (101.0 kB view details)

Uploaded Jul 20, 2026 CPython 3.12Windows x86-64

dvx_python-0.1.1-cp312-cp312-musllinux_1_2_x86_64.whl (210.3 kB view details)

Uploaded Jul 20, 2026 CPython 3.12musllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (112.7 kB view details)

Uploaded Jul 20, 2026 CPython 3.12manylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp312-cp312-macosx_11_0_arm64.whl (88.0 kB view details)

Uploaded Jul 20, 2026 CPython 3.12macOS 11.0+ ARM64

dvx_python-0.1.1-cp312-cp312-macosx_10_14_x86_64.whl (102.1 kB view details)

Uploaded Jul 20, 2026 CPython 3.12macOS 10.14+ x86-64

dvx_python-0.1.1-cp311-cp311-win_amd64.whl (103.4 kB view details)

Uploaded Jul 20, 2026 CPython 3.11Windows x86-64

dvx_python-0.1.1-cp311-cp311-musllinux_1_2_x86_64.whl (215.1 kB view details)

Uploaded Jul 20, 2026 CPython 3.11musllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (117.6 kB view details)

Uploaded Jul 20, 2026 CPython 3.11manylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp311-cp311-macosx_11_0_arm64.whl (89.9 kB view details)

Uploaded Jul 20, 2026 CPython 3.11macOS 11.0+ ARM64

dvx_python-0.1.1-cp311-cp311-macosx_10_14_x86_64.whl (103.8 kB view details)

Uploaded Jul 20, 2026 CPython 3.11macOS 10.14+ x86-64

dvx_python-0.1.1-cp310-cp310-win_amd64.whl (103.3 kB view details)

Uploaded Jul 20, 2026 CPython 3.10Windows x86-64

dvx_python-0.1.1-cp310-cp310-musllinux_1_2_x86_64.whl (215.6 kB view details)

Uploaded Jul 20, 2026 CPython 3.10musllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (117.5 kB view details)

Uploaded Jul 20, 2026 CPython 3.10manylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp310-cp310-macosx_11_0_arm64.whl (89.8 kB view details)

Uploaded Jul 20, 2026 CPython 3.10macOS 11.0+ ARM64

dvx_python-0.1.1-cp310-cp310-macosx_10_14_x86_64.whl (103.8 kB view details)

Uploaded Jul 20, 2026 CPython 3.10macOS 10.14+ x86-64

dvx_python-0.1.1-cp39-cp39-win_amd64.whl (103.6 kB view details)

Uploaded Jul 20, 2026 CPython 3.9Windows x86-64

dvx_python-0.1.1-cp39-cp39-musllinux_1_2_x86_64.whl (215.8 kB view details)

Uploaded Jul 20, 2026 CPython 3.9musllinux: musl 1.2+ x86-64

dvx_python-0.1.1-cp39-cp39-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl (117.8 kB view details)

Uploaded Jul 20, 2026 CPython 3.9manylinux: glibc 2.27+ x86-64manylinux: glibc 2.28+ x86-64

dvx_python-0.1.1-cp39-cp39-macosx_11_0_arm64.whl (89.9 kB view details)

Uploaded Jul 20, 2026 CPython 3.9macOS 11.0+ ARM64

dvx_python-0.1.1-cp39-cp39-macosx_10_14_x86_64.whl (104.0 kB view details)

Uploaded Jul 20, 2026 CPython 3.9macOS 10.14+ x86-64

File details

Details for the file dvx_python-0.1.1.tar.gz.

File metadata

Download URL: dvx_python-0.1.1.tar.gz
Upload date: Jul 20, 2026
Size: 2.3 MB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for dvx_python-0.1.1.tar.gz
Algorithm	Hash digest
SHA256	`9195699d4336e871318090ae4b23b5b77b67153cf25bf5c4c297881fe2f47105`
MD5	`71ed69bab11f261b84be55a8cc1906a1`
BLAKE2b-256	`4985014c98553923b23be4abf2e546c5171a28d9582bfc5c8453a0a3f09c01fe`

Algorithm	Hash digest
SHA256	`62e6fdafd65756161a3fe7b8194bffc74971247a32d0f3fbfbc96e7556981b2f`
MD5	`ed6429d204c64a011757355a33800b67`
BLAKE2b-256	`e0d14231f4238113cbb48321758a75583a95239a263304978c1fce406c0a8181`

Algorithm	Hash digest
SHA256	`58d0209bcc2fbc90d72d61da8c54a1af6838280f1ea205ad896a4b3cd3322a3a`
MD5	`2eeaa3ff22da9e02973bc0528138df79`
BLAKE2b-256	`5a16d93c3bcb3f395b6d3a3cc9ac3bcd17d4e75b7cb3341f3233e2c19ffa86e3`

Algorithm	Hash digest
SHA256	`4453f12c69c43d4dbdd882ae1f6e9a89af186c7b99ff6a5c328ed31da1a4640e`
MD5	`697f094b8b6bd7dedce335b1c900a8ba`
BLAKE2b-256	`7271a81390e1507eb51a98c75149746bfa4f5bcc027db7e88be17db76bb3975a`

Algorithm	Hash digest
SHA256	`cb35e76168a752a6aac6c5d1b91abb50a9b3a39ca35828971fb4082cabcd1b3c`
MD5	`fb861efa005406687e2006a73b8b11a3`
BLAKE2b-256	`852b81266ca9098b7ad0507ae01a586aa44afdaac1dbcba7252cc36da4f52465`

Algorithm	Hash digest
SHA256	`be6821bf80c602fcbc4ddece04a2a3077b7efbac9cd24d57b356b6d7d07b7560`
MD5	`e38fcf9041fee1b03d3799fe54060e51`
BLAKE2b-256	`032c8b241f207fefe9198e3072c2a51567469ce4840e1f1c015d1799d8b376c3`

Algorithm	Hash digest
SHA256	`e895bd0906c9def5b4b0dda135e4e0bd5f4a646ce418fc2e723a83bb50be9ecd`
MD5	`496847a4ab6e8b81ea215a2345614727`
BLAKE2b-256	`49c57856ca263a83eac988f33a6555f8d61045d072a3fdf862e1f2ed963e3407`

Algorithm	Hash digest
SHA256	`d268d60d04334880c5224f73481de6e1fc83bc8ede4a3efbc6852d27c119740c`
MD5	`00f6edc2ba21b328ebe674cb3084d7bf`
BLAKE2b-256	`1e476e4b3383b51a499679ad82b06fcdb344ed7f92e104966e59d2bbbce3a878`

Algorithm	Hash digest
SHA256	`66b57761a1ff37fa42712606b8deac1591193f6734fb5df85f1f22d6f1c55d3f`
MD5	`41d744fb0ac6ec933b710985ff1bf0b8`
BLAKE2b-256	`c261918eb0210102bf6496942b1e9c9eea0e2cf230c0260864304f01992f3e0d`

Algorithm	Hash digest
SHA256	`575fa5e9d0ad431c0fc426199db9eadc4c4e84fa703279f2d4b19ed1b3202a56`
MD5	`f00f7959a07cab732e04ac934cba6dcf`
BLAKE2b-256	`84ddbbbef8a8661c49ccd949714539287b0f2002d760b3746dc63802f5a8d6ce`

Algorithm	Hash digest
SHA256	`ae9c6cdf66727ff031dda087258fc263f4606676e0af6666aa63374b2e7e9942`
MD5	`f32328ac134269c35b0a2c935584aea4`
BLAKE2b-256	`56eabc5c647c2b0b64554fb6e9d9bfd34a0862c6661643035628a30c6d614495`

Algorithm	Hash digest
SHA256	`c245c25c837b24adf245fbb2954a94ff2e60db03d2be4544002e25f6afb00485`
MD5	`cdf5e836c449d962ec0afe987ba6f845`
BLAKE2b-256	`ddec1ebca5612bb7778559ce0502c373be607762ad6dc36f25e84e0eca03bb45`

Algorithm	Hash digest
SHA256	`7b0b8130fa5cf3d7071a205c38cada98215cddcfc978cbd539f7ede5b52ad649`
MD5	`0881be72068033bd2e62ad03e71f61d0`
BLAKE2b-256	`99c787bd290104d5943d4d0b37f4eda4580e1edd58f2ac146e000b2a9720b49d`

Algorithm	Hash digest
SHA256	`d8d679406e61b03ba6239943d4d19c32be5994b8ad32e87d3c138c730456b0a9`
MD5	`811c683e150cc0965efdfca48b95d895`
BLAKE2b-256	`aed929eee2963696eaf87288c3d8aaf38dde422a3730cc66babd8a150bb2da75`

Algorithm	Hash digest
SHA256	`043dc032bfcca782071974d77931b674e84412ecef37a924eeaa24ee9b773cae`
MD5	`e1bf7331ac65d781a7d41263b5953951`
BLAKE2b-256	`a1f3832507f9ba90621372ee024b3c33f905cf4e496184bd142fab41fac5ccf4`

Algorithm	Hash digest
SHA256	`bc58b9f13eea50bab8c8889d083287a74b7f0d57a7725d677fcb77ae33c4aba4`
MD5	`6499f327c7202f9db973d9f4b995084d`
BLAKE2b-256	`775e444a6912ff48a1e98f061fc19f60219905d7f0ba68a9c89a36b13d4f7faf`

Algorithm	Hash digest
SHA256	`13fd8046c8ec97728b034d740e38cfba69ab076a052a79dbe59d8dbb11148980`
MD5	`2b01cf4e8562a90bb7d7c29812982452`
BLAKE2b-256	`31ed0ece56444e8e0159f99077a1656b3a5cd47e7512686b055627190a245850`

Algorithm	Hash digest
SHA256	`9bfe85819929ba543335ff9572089e56f6094b11066540ea7acb369a9da1584a`
MD5	`a588b19db5be571d150ff5759f34e294`
BLAKE2b-256	`7a4ea3995571b68ce9a2272eb839b60aa978f3b12c33a767c3eb964dfdae9b4b`

Algorithm	Hash digest
SHA256	`399f4cac2d2b9acd68bcd85e94311faae494dd94b95d1d2a4890833b51588621`
MD5	`4a4748501c0fa2a9710ef878523d984e`
BLAKE2b-256	`5f80db4a53be26925a9fa370a805a17cad7e5dfd596d3c948e2726310ac1d424`

Algorithm	Hash digest
SHA256	`f238ab712a5358554a8b9c81bcd293ad27b2ed14fdd0b2529e56dad861e6eaba`
MD5	`dced8a9d9367308816eab1086a1f7ac5`
BLAKE2b-256	`1fe6e8b1ba8cafc2751fcf3d52b0a8ee5ab5952132378b3e364bdf2c4c2d5040`

Algorithm	Hash digest
SHA256	`958084583c9cb5d600043077e1b671a824ac4095f028f7b9d410908c780f58f8`
MD5	`9a55f77f90728d7c4e234f2c754be00e`
BLAKE2b-256	`0785095c9b87bc2a8583ef54f7a492e464f619688fdf82e90a16f89a8a5b4d23`

dvx-python 0.1.1

Differentiable Voxelization of Surface Representations

About

Getting Started

Optional: Test the Installation

Usage

Conventions for Coordinates and Data Layout

License and Copyright

Citation

Project links

Key dates

1 maintainer

Credits

License

Requires

Provides Extra

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

Algorithm	Hash digest
SHA256	`438ca6d91520a252643c000191dbe7f21f1173b2ac6645601b6db971ccacc63c`
MD5	`5dd88b7f914f2e09ba64fc32ca36b8c1`
BLAKE2b-256	`55946c899d6c11b95c4a161819bbcc445a32bc5fd2f3725423609317936bb90e`

Algorithm	Hash digest
SHA256	`646fe7f1ff17564500f4f6754d0197887a7997a24a3daab5c02ee9e7cf8ef524`
MD5	`8e52b9fab5709be1841d76726e3b271d`
BLAKE2b-256	`dbac4906ce93cd16a8162e3054ca41513401849f6cda17bf70cf7216918b481c`

Algorithm	Hash digest
SHA256	`4fea41055d757213bd7bbad0e163d1bfb25a3cd0b73778d78823cb0e0a2d3fa1`
MD5	`25bbdd680cc992661ca7d20f18933f7e`
BLAKE2b-256	`f62b91af80c099a0a6696aa5bc1204e78d1149cf74ec47d75c72dda93d5daefb`

Algorithm	Hash digest
SHA256	`ac8161d6d1a03f6c73d31a9c0538bb6c968742dddf3464c67edc603e6f59750d`
MD5	`68d56a7e1d699cbfd561f1e2cbb7239d`
BLAKE2b-256	`e9c3d570bce157db70f053c4f638d591132ee06be2b63fd69719b9a319a0bc6f`

Algorithm	Hash digest
SHA256	`7ed3a32fab23f13dd974a2d6b475e6e36373ea2cdf59d7876f0a142f5936e721`
MD5	`5c114fe2e4e6cc4b6f2664edda7827ab`
BLAKE2b-256	`8db61eb243918d2e1ffcf4cc10a169a3979430f14ee070931df8884f70523a62`

Algorithm	Hash digest
SHA256	`a72ea9ad2cb52ee150b0dca2d6d4ddf94ea1c7e2cceb2d35cc576305690be421`
MD5	`1be7c063662c8bd529b01db8ca791ea3`
BLAKE2b-256	`c241f6310b94d06329a6e12064ef0f2b64f0ccfaa693666b61a9eb896959c0d9`

Algorithm	Hash digest
SHA256	`aa543e971d90dc3edd298f0ea8193778b27a1a1686642b2a172d9505dee2cda1`
MD5	`28e066ca0600d205bdf66ba8575e359a`
BLAKE2b-256	`e10a6c9606bb82f1942233d2260695f0e02be29fcf3eb07151afa73849aa73fb`

Algorithm	Hash digest
SHA256	`a1280842589cdee506bcf544b0c9a0003457b4bb3199b14d5c47015c38f86f91`
MD5	`6d2e12e0cc1da56d5ab656f558c68d32`
BLAKE2b-256	`15fa95e64c064119b89c551ff0ce8d0d80f613dbfe79e612fc11fc435ab8bb75`

Algorithm	Hash digest
SHA256	`64428c5eeed09bd51c5d00af3f930e348f92dc0e37ba6460c1c6b63c71907d47`
MD5	`f736659f35b39644adad737def62982a`
BLAKE2b-256	`1819047dae6b1cadd117aff85da0d45ec69452102e01c180090020b5f1847e4c`

Algorithm	Hash digest
SHA256	`d100440732069efc4bedd1fd82795f990784e02aa1c722ca8427694782edd389`
MD5	`217284caae6587a738cc9ad0613a94ff`
BLAKE2b-256	`4bf4aba3fb9ad5202a894f9900e25c5448a036e8e2b96ef27ba74b1115dbca31`