Datoviz: high-performance GPU scientific data visualization C/C++ library¶

[Installation]  [Usage]  [User guide]  [Examples]  [API reference]  

⚡️ Datoviz is a cross-platform, open-source, high-performance GPU scientific data visualization library written inC/C++ on top of theKhronos Vulkan graphics API and theglfw window library. It provides raw ctypes bindings inPython 🐍. In the long term, Datoviz will mostly be used as aVisPy 2.0 backend.

Designed for speed, visual quality, and scalability to datasets comprising millions of points, it supports 2D/3D interactive rendering and minimal GUIs viaDear ImGui.

⚠️ Warning: Although Datoviz has been years in the making, it is still in itsearly stages and would greatly benefit from increasedcommunity feedback, particularly concerning package and hardware compatibility. The API is still evolving, so expect regular (though hopefully minimal)breaking changes for now. The current version isv0.2. The previousv0.1 version is now deprecated.

🕐 Roadmap. In the medium term: increasing OS and hardware compatibility, providing more visuals, interactivity patterns, and GUI controls. In the long term: picking, custom visuals and shaders, nonlinear transforms, WebGPU/WebAssembly compatibility, integration with IPython, Jupyter and Qt.

🖼️ Screenshots from the v0.1 version¶

Credits: mouse brain volume:Allen SDK. France:Natural Earth. Molecule:Crystal structure of S. pyogenes Cas9 from PDB (thanks to Eric for conversion to OBJ mesh). Earth:Pixabay. Raster plot: IBL. 3D human brain:Anneke Alkemade et al. 2020, thanks to Pierre-Louis Bazin and Julia Huntenburg.

✨ Features¶

• 📊 High-quality antialiased 2D visuals: markers, lines, paths, glyphs
• 🌐 3D visuals: meshes, volumes, volume slices
• 🌈 150 colormaps included (from matplotlib, colorcet, MATLAB)
• 🖱️ High-level interactivity: pan & zoom for 2D, arcball for 3D (more later)
• 🎥 Manual control of cameras: custom interactivity
• 𓈈 Figure subplots (aka "panels")
• 🖥️ Minimal GUIs usingDear ImGui

List of visuals¶

Work in progress¶

These features are currently planned forv0.3:

• ➕ Axes: ticks, grids, labels
• 🎨 Colorbars
• 🖱️ More interactivity patterns
• 📖 More documentation

Future work¶

These features are currently planned forv0.4 and later:

• 📐 More visuals: arrows, polygons, planar straight-line graphs (PSLG), histograms, areas, graphs
• 🎯 Picking
• 📈 Nonlinear transforms
• 🖌️ Dynamic shaders
• 🌐 WebGPU/WebAssembly compatibility
• 🧮 Compute shaders
• 🐍 IPython, Jupyter, Qt integration

🛠️ Installation instructions¶

Requirements:

• A supported OS (Linux, macOS 12+, Windows 10+)
• A Vulkan-capable graphics chipset (either integrated or dedicated graphics process unit)
• Python and NumPy

Note: You no longer need to install the Vulkan SDK or to manually build the library. Precompiled wheels for Linux, Windows, and macOS have been uploaded to PyPI.

pipinstalldatoviz

🚀 Usage¶

Simple scatter plot example (points with random positions, colors, and sizes) in Python, which closely follow the C API.

importnumpyasnpimportdatovizasdvzapp=dvz.app(0)batch=dvz.app_batch(app)scene=dvz.scene(batch)figure=dvz.figure(scene,800,600,0)panel=dvz.panel_default(figure)dvz.panel_panzoom(panel)visual=dvz.point(batch,0)n=100_000dvz.point_alloc(visual,n)pos=np.random.normal(size=(n,3),scale=.25).astype(np.float32)dvz.point_position(visual,0,n,pos,0)color=np.random.uniform(size=(n,4),low=50,high=240).astype(np.uint8)dvz.point_color(visual,0,n,color,0)size=np.random.uniform(size=(n,),low=10,high=30).astype(np.float32)dvz.point_size(visual,0,n,size,0)dvz.panel_visual(panel,visual,0)dvz.scene_run(scene,app,0)dvz.scene_destroy(scene)dvz.app_destroy(app)

Check out theexamples documentation for more usage examples.

📚 Documentation¶

• 📖 User guide
• 🐍 Examples
• 📚 API reference
• 🏛️ Architecture overview
• 🏗️ Build instructions
• 👥 Contributors instructions
• 🛠️ Maintainers instructions

🕰️ History and current status¶

In2012, developers of various GPU scientific visualization libraries (Galry, Glumpy, pyqtgraph, visvis) collaborated to createVisPy, an OpenGL-based scientific visualization library for Python.

In2015,Vulkan, the successor to OpenGL, was announced by Khronos,sparking the idea of a future Vulkan-based visualization library.

In2019,Cyrille Rossant, one of the original VisPy developers, began experimenting with Vulkan.

In2021, thefirst experimental version of Datovizv0.1 was released. This initial release laid the groundwork for further development.

Over the next three years, the technology matured, aided by aChan Zuckerberg Initiative (CZI) grant awarded to VisPy in2021.

In2024, Datovizv0.2 is released. This version is redesigned from the ground up to enhance modularity and stability, ensuring it can keep pace with the continuous advancements in GPU hardware and graphics rendering APIs.It features a modular architecture that will allow the porting of Datoviz technology to non-Vulkan environments, such as WebGPU-enabled web browsers (thanks to a secondCZI grant).

Datoviz is closely related toVisPy, as it is being developed by one of the VisPy cofounders. VisPy 2.0, initiated by Cyrille Rossant and Nicolas Rougier, will offer a high-level scientific API on top of Datoviz, matplotlib, and other renderers via a common medium-level visualization layer called "graphics server protocol (GSP)".

The long-term vision is for high-performance GPU-based 2D/3D scientific visualization to be uniformly available across multiple platforms, environments (desktop, web, cloud-based remote visualization), and programming languages (C/C++, Python, Julia, Rust, etc.).

🤝 Contributing¶

See thecontributing notes.

📄 License¶

See theMIT license.

🙏 Credits¶

Datoviz is developed byCyrille Rossant at theInternational Brain Laboratory, a consortium of neuroscience research labs around the world.

It is funded notably byChan Zuckerberg Initiative'sEssential Open Source Software for Science program.