Show HN: Datoviz – Vulkan-based GPU scientific visualization (C/C++/Python)

[rossant]

https://github.com/datoviz/datoviz

I'm excited to announce the release of Datoviz 0.2.0, an open-source, high-performance GPU scientific visualization library built on Vulkan. It targets the interactive visualization of large 2D/3D datasets. This version includes tentative precompiled Python wheels for Linux, macOS (ARM and Intel), and Windows.

Datoviz is a key part of the CZI-funded Vispy 2.0 project and will serve as its main GPU backend. Datoviz provides core GPU visualization capabilities while VisPy 2.0 will provide high-level plotting functionality (a bit similar to NumPy vs SciPy).

What I'm looking for from the community:

1. Compatibility feedback: I'd appreciate quick feedback on how these precompiled Python wheels perform across different operating systems and graphics hardware.

2. Library feedback: Datoviz is still in its early stages and actively evolving. The API is subject to change, and I'd appreciate any feedback on its functionality and design.

Please report issues on GitHub.

Since the initial 2021 v0.1 release, the underlying technology has matured significantly. The internal architecture is now more robust and modular, paving the way for support for other rendering technologies like WebGPU and WebAssembly in the medium term.

While still somewhat limited, the focus is on ensuring stability, performance, and visual quality. Available visuals include points, markers, line segments, paths, glyphs, images, spheres, 3D meshes, and basic volume rendering. Some important features, like axes and picking, are planned for v0.3 and later.

Please give it a try, report any issues on GitHub, and feel free to ask questions. You're also welcome to contribute. I'm looking forward to your feedback!

[radarsat1]

Is this similar in spirit to VTK? (An API I've not always found super intuitive so interested in alternatives.)

Or is this better thought of as a powerful matplotlib alternative?

[rossant → radarsat1]

Yes, it's somewhat similar in spirit to VTK, though VTK is obviously much more mature and feature-rich (it was created in 1993).

VTK offers more advanced capabilities (computational geometry algorithms, image processing, advanced volume rendering...) and can be seen as a comprehensive all-in-one toolbox for 3D scientific visualization.

Datoviz, on the other hand, is much more lightweight, focusing solely on rendering. A significant effort in Datoviz has gone into delivering high-quality, high-performance 2D visualization, whereas VTK's strength lies in 3D rendering.

Datoviz also delegates nearly all computational tasks to other libraries. There are a few exceptions: for instance, Datoviz will likely soon include basic computational geometry algorithms, such as polygon triangulation (using ear clipping, for example), necessary for the upcoming polygon visual. However, these will be limited to the most essential types of visual elements.

Here are a few reference points for a rough comparison:

- VTK: millions of lines of C/C++ code, with compilation potentially taking tens of minutes (?).

- Datoviz: 35,000 lines of C/C++ code, with fresh compilation taking only 15 seconds on a recent MacBook Air.