Frequently Asked Questions#
General#
Q: What is the difference between kaolin and kaolin-wisp?
A: kaolin is a 3d deep learning library in the broader sense. It contains many useful operations for handling differentiable 3D graphics, like a mesh toolkit, a cameras module, structured pointcloud (e.g. octrees), differentiable renderers, and more. It builds on, and extends PyTorch with useful operations for handling general 3D structures.
kaolin-wisp (abbreviated as just “wisp”), is dedicated for neural fields research. It uses kaolin’s components to build full pipelines
Q: Why is the library called “wisp”?
A: Our library is named after the atmospheric ghost light, will-o’-the-wisp, which are volumetric ghosts that are harder to model with common standard geometry representations like meshes. We provide a multiview dataset of the wisp as a reference dataset for a volumetric object. We also provide the blender file and rendering scripts if you want to generate specific data with this scene, please refer to the readme.md for greater details on how to generate the data.
Q: Is kaolin-wisp affiliated with Instant-NGP?
A: Both projects come from NVIDIA research labs, but are purposed differently. Instant-NGP is a well-optimized codebase, which relies on fused CUDA kernels for training high quality Neural Radiance Fields. Wisp on the other hand takes a flex approach and puts emphasis on modularity and the ability to customize code. As a design choice, Wisp relies on PyTorch, and where appropriate, CUDA kernels are used to optimize critical sections of the code.
If you’re using NeRF as a black-box, Instant-NGP is a great choice.
If you require customized code, e.g. “hacking” into the neural field or training structure, Wisp is the way to go.
Q: About project’s roadmap?
A: Short-mid term Milestones are logged on github.
Licensing & Contributions#
Q: Is kaolin-wisp an open source?
A: Wisp is licensed under the NVIDIA Source Code License. That means you are welcome, and encouraged to use it for research! The exact terms can be reviewed on the License page.
Q: Are there any limitations to hosting my pipelines on kaolin-wisp’s github?
A: Not at all! We welcome new pipeline contributions, whether as apps or examples. Credit will be given due on the README page included with the app. Please include a short sample script which allows us to test your pipeline with our CI.
Q: My project forks kaolin-wisp, can I still link it back?
A: Sure, feel free to contact us on Github and we will link it back.
Q: How do I contribute to the codebase?
A: Contributions are expected in the form of pull requests.
Common Issues#
Q: Does wisp support headless rendering on remote machines?
A: If your machine is not equipped with a GPU or display, you may still run a simplified renderer via Jupyter Notebook.
The full interactive visualizer is limited to machines equipped with an NVIDIA GPU and a display. The minimum requirements come from usage of OpenGL.
Q: Can I run wisp on Windows?
A: Windows support is provided by external contributions from the community. Note that our CI does not track Windows status.
Q: Do I require a GPU to run wisp?
A: Both kaolin and kaolin-wisp rely on CUDA kernels, which require an NVIDIA GPU.
Q: I pulled the latest on main branch and now can’t run the NeRF app training code.
A: Since NeRF is moving at a very fast pace, our changes to the main branch may break backwards compatability from time to time.
When that happens, we always leave a note within the main page (see: “Latest Updates”).
In addition, also keep the stable branch up to date with the latest tagged version, to ensure users have a point of reference.
Reproducibility#
Q: I’m unable to reproduce the results of paper X.
A: The metrics obtained from wisp apps are provided in their respective README pages (see NeRF for example). Note that wisp is meant to be used as a library of components: we do not guarantee an identical reimplementation of papers, The metrics obtained by our reference implementations should have comparable, but not identical metrics.
Q: Are models trained with wisp compatible between versions?
A: We track certain configurations and metrics with our CI, to ensure they don’t degrade between versions. If you spot something unusual, please notify us on Github.