pygfx

A render engine, inspired by ThreeJS, but for Python and targeting Vulkan/Metal/DX12 (via wgpu).

Introduction

This is a Python render engine build on top of WGPU (instead of OpenGL).

We take a lot of inspiration from ThreeJS, e.g.:

• Materials and Geometry are combined in world objects.
• Decoupled cameras and controllers.
• The code for the render engines is decoupled from the objects, allowing multiple render engines (e.g. wgpu and svg).

Further we aim for a few niceties:

• Proper support for high-res screens.
• Builtin anti-aliasing.
• Custom post-processing steps.
• Support for picking objects and parts within objects.
• (approximate) order-independent transparency (OIT) (not implemented yet).

WGPU is awesome (but also very new)

Working with the WGPU API feels so much nicer than OpenGL. It's well defined, no global state, we can use compute shaders, use storage buffers (random access), etc.

Fair enough, the WGPU API is very new and is still changing a lot, but eventually it will become stable. One of the biggest downsides right now is the lack of software rendering. No luck trying to run wgpu on a VM or CI.

Because of how Vulkan et. al. work, the WGPU API is aimed at predefining objects and pipelines and then executing these. Almost everything is "prepared". The main reasoning for this is consistency and stable drivers, but it also has a big advantage for us Pythoneers: the amount of code per-draw-per-object is very limited. This means we can have a lot of objects and still be fast.

As an example, see collections_line.py: drawing 1000 line objects with 30k points each at 57 FPS (on my laptop).

How to build a visialization

See also the examples, they all do something like this:

• Instantiate a renderer and a canvas to render to.
• Create a scene and populate it with world objects.
• Create a camera (and maybe a control).
• Define an animate function that calls: renderer.render(scene, camera)

On world objects, materials, and geometry

There are a few different world object classes. The class defines (semantically) the kind of object being drawn, e.g. Line, Image, Mesh, Volume. World objects have a position and orientation in the scene, and can have children (other world objects), creating a tree. World objects can also have a geometry and/or a material.

The geometry of an object defines its base data, usually per-vertex attributes such as positions, normals, and texture coordinates. There are several pre-defined geometries, most of which simply define certain 3D shapes.

The material of an object defines how an object is rendered. Usually each WorldObject class has one or more materials associated with it. E.g. a line can be drawn solid, segmented or with arrows. A volume can be rendered as a slice, MIP, or something else.

Installation

pip install -U pygfx

Or, to get the latest from GitHub:

pip install -U https://github.com/pygfx/pygfx/archive/main.zip

Current status

Under development, many things can change.

Testing examples

There are two types of tests for examples included with pygfx:

Type 1: Checking if examples can run

When running the test suite, pytest will run every example in a subprocess, to see if it can run and exit cleanly. You can opt out of this mechanism by including the comment # run_example = false in the module.

Type 2: Checking if examples output an image

You can also (independently) opt-in to output testing for examples, by including the comment # test_example = true in the module. Output testing means the test suite will attempt to import the renderer instance global from your example, and call it to see if an image is produced.

To support this type of testing, ensure the following requirements are met:

• The WgpuCanvas class is imported from the wgpu.gui.auto module.
• The renderer instance is exposed as a global in the module.
• A rendering callback has been registered with renderer.request_draw(fn).

Reference screenshots are stored in the examples/screenshots folder, the test suite will compare the rendered image with the reference.

Note: this step will be skipped when not running on CI. Since images will have subtle differences depending on the system on which they are rendered, that would make the tests unreliable.

For every test that fails on screenshot verification, diffs will be generated for the rgb and alpha channels and made available in the examples/screenshots/diffs folder. On CI, the examples/screenshots folder will be published as a build artifact so you can download and inspect the differences.

If you want to update the reference screenshot for a given example, you can grab those from the build artifacts as well and commit them to your branch.