Repository files navigation

[Installation]  [Usage]  [User guide]  [Gallery]  

⚡️ Datoviz is an open-source, cross-platform,high-performance rendering library for scientific data visualization making extensive use of the graphics processing unit (GPU).

Up to10,000x faster than matplotlib, it delivers highly efficienthigh-quality GPU rendering of 2D and 3D graphical primitives—markers, paths, images, text, meshes, volumes, and more—that scale to millions of elements. Datoviz also supportsgraphical user interfaces (GUIs) for interactive visualization.

Built from the ground up with performance in mind, Datoviz is written primarily inC andC++, leveraging theKhronos Vulkan graphics API. It offers a C API, low-level Python bindings viactypes, and a higher-level, idiomaticPython API 🐍.

Written by one of the original creators ofVisPy, a GPU-based Python scientific visualization library, Datoviz aims to serve as the default backend for the upcomingVisPy 2.0.

The library is lightweight with minimal dependencies: mostly Vulkan,GLFW for windowing, andDear ImGui for GUIs.

• 📊 2D visuals: antialiased points, markers, line segments, paths, text, images
• 📈 2D axes
• 🌐 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")
• 🖥️ GUIs usingDear ImGui

Datoviz doesnot use high-level plotting functions likeplot(),scatter(), orimshow(). Instead, it exposes flexible visual primitives—markers, images, meshes, and more—that can be added to a scene and customized directly with data. This approach offers fine-grained control and high performance, while remaining concise enough for quick plots.

Ahigher-level plotting interface is under development as part ofVisPy 2.0, which will use Datoviz as its rendering backend. An intermediate layer calledGSP (Graphics Specification Protocol) will provide a backend-agnostic API for declarative plotting.

Matplotlib is the gold standard for static, publication-quality figures, with a mature, feature-rich API. But it's not optimized for interactivity or large datasets.

Datoviz, in contrast, is a younger, GPU-based library designed for fast, interactive visualization. It lacks Matplotlib’s high-level API but excels at rendering millions of points in real time.

UseMatplotlib for polished static plots,Datoviz for responsive data exploration.

The figure below shows a preliminaryperformance benchmark comparing Datoviz and Matplotlib on an interactive 2D scatter plot, with increasing numbers of points and a simulated zoom interaction (on a high-end Linux desktop PC):

In this benchmark,Datoviz outperforms Matplotlib by a factor of up to 10,000×, maintaining interactive framerates even with millions of points. Matplotlib, in contrast, becomes sluggish or fails entirely as the dataset size increases.

Datoviz is lighter, simpler, and easier to install than VTK. It compiles in seconds, has minimal dependencies, and focuses exclusively on real-time rendering of visual primitives. Unlike VTK, it does not include data file I/O (except minimal test loaders like.obj), computational geometry, or data processing.

VTK is a powerful, heavyweight toolkit for 3D visualization, simulation, and scientific computing workflows. In contrast, Datoviz is designed for fast, high-quality 2D and 3D rendering.

The current version is v0.3.xThis release introduces major updates over v0.2, including 2D axes and a new Pythonic API.

Looking ahead, the upcoming v0.4 release (2026) will focus on foundational improvements to the low-level engine, paving the way for the following key features in future versions:

• 🧊 Correct transparency in 3D mesh and volume rendering
• ✨ Multisample anti-aliasing (MSAA)
• 🎯 Object picking
• 📈 Nonlinear coordinate transforms
• ⚡ CUDA interoperability
• 🧮 Vulkan compute shaders (similar to CUDA kernels)
• 🖌️ Dynamic and customizable shaders
• 🎛️ Combined GPGPU compute and graphics workflows
• 🔗 GPU memory sharing across visuals
• 🐍 IPython integration
• 🖥️ Qt backend support
• 🌐 WebGPU backend

• Supported operating systems:

  • Linux x86_64 (e.g. Ubuntu 22.04 or later, as the Linux build requires glibc 2.34+)
  • macOS 12 or later (x86_64 and arm64, i.e. Apple silicon M1–M4)
  • Windows 10 or later (x86_64)

• Python environment:

  • Python 3.8+ (64-bit)
  • NumPy
  • requests
  • platformdirs

• Up-to-date graphics drivers with Vulkan support (most integrated or dedicated GPUs from the last ten years are supported)

pip install datoviz

This installs a Python wheel that includes the C library, precompiled for your system.

If the installation fails, you may need tobuild from source oropen an issue to request support for your configuration.

Nightly builds are automatically generated from thedev branch viaGitHub Actions. These builds contain the latest experimental features and bug fixes that have not yet been released.

You can find the latest builds on theNightly Release Page.Each wheel file is tagged by platform (e.g.manylinux,macosx,win_amd64) and build date (devYYYYMMDD).

Here’s a simple 2D scatter plot example with axes in Python, displaying points with random positions, colors, and sizes.

importnumpyasnpimportdatovizasdvzn=1000x=np.random.normal(scale=0.2,size=n)y=np.random.normal(scale=0.2,size=n)color=np.random.randint(size=(n,4),low=100,high=240,dtype=np.uint8)color[:,3]=255size=np.random.uniform(low=10,high=30,size=n)app=dvz.App(background='white')figure=app.figure(800,600)panel=figure.panel()xmin,xmax=-1,+1ymin,ymax=-1,+1axes=panel.axes((xmin,xmax), (ymin,ymax))visual=app.point(position=axes.normalize(x,y),color=color,size=size,)panel.add(visual)app.run()app.destroy()

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

Datoviz builds on more than a decade of open-source GPU-based scientific visualization work:

• 2012 — Developers of several GPU visualization libraries (Galry, Glumpy, pyqtgraph, visvis) joined forces to createVisPy, a high-performance OpenGL-based visualization library for Python.

• 2015 — TheVulkan API was introduced as a modern low-level successor to OpenGL,inspiring early ideas for a future visualization library built on Vulkan.

• 2019 —Cyrille Rossant, a VisPy cofounder, began experimenting with Vulkan as the foundation for a new graphics engine tailored to scientific visualization.

• 2021 — The first experimental release ofDatoviz v0.1 laid the foundation for future development. That same year, aChan Zuckerberg Initiative (CZI) grant to VisPy helped support the ecosystem.

• 2024 — Datovizv0.2 was released with a full redesign. It introduced a modular architecture focused on stability and extensibility, paving the way for cross-platform rendering (including WebGPU) with support from a secondCZI grant.

• 2025 — Datovizv0.3 added 2D axes, a more Pythonic API, and core improvements in usability and flexibility.

Datoviz remains closely tied toVisPy and is being developed by one of its original authors. As part of theVisPy 2.0 initiative (led by Cyrille Rossant and Nicolas Rougier, two of the original creators of VisPy), Datoviz will act as a low-level backend beneath a unified scientific visualization layer called theGraphics Server Protocol (GSP), enabling frontends to target multiple renderers (Datoviz, Matplotlib, etc.).

The long-term vision is to enable high-performance 2D/3D scientific visualization across platforms (desktop, web, cloud) and languages (C/C++, Python, Julia, Rust).

See thecontributing notes.

See theMIT license.

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

Datoviz is supported by:

• Wellcome Trust
• Simons Foundation
• Chan Zuckerberg Initiative, through theEssential Open Source Software for Science program

• Khronos Group (2025).Datoviz: Ultra-fast, high-performance GPU scientific visualization library built on Vulkan
• Rossant, C. & Rougier, N. (2021).Datoviz: A GPU Scientific Visualization Library Based on Vulkan.2021 IEEE Visualization Conference (VIS).
• Rossant, C. (2015).A compiler for data visualization
• Campagnola, L., Klein, A., Larson, E., Rossant, C., & Rougier, N. P. (2015, July).VisPy: harnessing the GPU for fast, high-level visualization. In Proceedings of the 14th Python in Science Conference.
• Rougier, N. P. (2014).Antialiased 2d grid, marker, and arrow shaders. Journal of Computer Graphics Techniques, 3(4), 52.
• Rougier, N. P. (2013).Shader-based antialiased dashed stroked polylines. Journal of Computer Graphics Techniques, 2(2), 91-107.
• Rougier, N. P. (2013).Higher quality 2D text rendering. Journal of Computer Graphics Techniques, 2(1), 50-64.