simpleimageio 1.5.1

A very simple Python wrapper to read and write various HDR and LDR image file formats.

Simple Image IO

A lightweight C# and Python wrapper to read and write RGB images from / to various file formats. Supports .exr (with layers) via tinyexr and a number of other formats (including .png, .jpg, and .bmp) via stb_image and stb_image_write. A subset of TIFF can be read and written via tinydngloader. We also implement our own importer and exporter for PFM. In addition, the package offers some basic image manipulation functionality, error metrics, and tone mapping.

The C# wrapper further offers utilities for thread-safe atomic splatting of pixel values, and sending image data to the tev viewer via sockets. It also contains a very basic wrapper around Intel Open Image Denoise.

The Nuget package contains prebuilt binaries of the C++ wrapper for x86-64 Windows, Ubuntu, and macOS (.github/workflows/build.yml). The Python package is set up to automatically download an adequate CMake version and compile the C++ code on any platform.

Except for the optional Intel Open Image Denoise, all dependencies are header-only and unintrusive, so this library should work pretty much anywhere without any hassle.

Usage example (C#)

The following creates a one pixel image and writes it to various file formats:

RgbImage img = new(width: 1, height: 1);
img.SetPixel(0, 0, new(0.1f, 0.4f, 0.9f));
img.WriteToFile("test.exr");
img.WriteToFile("test.png");
img.WriteToFile("test.jpg");

Reading an image from one of the supported formats is equally simple:

RgbImage img = new("test.exr");
Console.WriteLine(img.GetPixel(0, 0).Luminance);

The pixel coordinate (0,0) corresponds to the top left corner of the image. Coordinates outside the valid range are clamped automatically; no error is raised. The framework also offers a MonochromeImage with a single channel per pixel. Further, the base class ImageBase can be used directly for images with arbitrary channel count (RgbImage and MonochromeImage only add some convenience functions like directly returning an RgbColor object).

As an added bonus, the C# wrapper can connect to the tev HDR viewer and directly display image data via sockets. The following example generates a monochrome image and sends it to tev:

TevIpc tevIpc = new(); // uses tev's default port on localhost

// Create the image and initialize a tev sync
MonochromeImage image = new(width: 20, height: 10);
tevIpc.CreateImageSync("MyAwesomeImage", 20, 10, ("", image));

// Pretend we are a renderer and write some image data.
image.SetPixel(0, 0, val: 1);
image.SetPixel(10, 0, val: 2);
image.SetPixel(0, 9, val: 5);
image.SetPixel(10, 9, val: 10);

// Tell the TevIpc class to update the image displayed by tev
// (this currently retransmitts all pixel values)
tevIpc.UpdateImage("MyAwesomeImage");

Usage example (Python)

The following creates a one pixel image, writes it to various file formats, reads one of them back in, and prints the red color channel of the pixel. The result is then sent to the tev HDR viewer via sockets (modified version of https://gist.github.com/tomasiser/5e3bacd72df30f7efc3037cb95a039d3).

import simpleimageio as sio
sio.write("test.exr", [[[0.1, 0.4, 0.9]]])
sio.write("test.png", [[[0.1, 0.4, 0.9]]])
sio.write("test.jpg", [[[0.1, 0.4, 0.9]]])
img = sio.read("test.exr")
print(img[0,0,0])

with sio.TevIpc() as tev:
    tev.display_image("image", img)
    tev.display_layered_image("layers", { "stuff": img, "morestuff": img })

In Python, an image is a 3D row-major array, where [0,0,0] is the red color channel of the top left corner. The convention is compatible with most other libraries that make use of numpy arrays for image representation, like matplotlib.

Flip books for Jupyter and web

Both, the Python and the .NET library can generate an interactive HTML viewer to display and compare images visually by flipping between them. See FlipBookExample.dib for an example with .NET interactive and C#, FlipBookExample.fsx for a static webpage generator with F#, or flipbook.ipynb for a Jupyter notebook with Python.

Building from source

If you are on an architecture different from x86-64, you will need to compile the C++ wrapper from source. Below, you can find instructions on how to accomplish that.

Dependencies

All dependencies except OIDN are header-only and included in the repository. Building requires

• a C++20 compiler
• CMake
• .NET 6.0 (or newer)
• Python ≥ 3.6

Building the C# wrapper - quick and simple

pwsh ./make.ps1

Downloads precompiled binaries for Open Image Denoise, copies them to the correct directory, builds the C++ code and then builds and tests the C# wrapper.

Building the C# wrapper - manually

Build the C++ low level library with CMake:

mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
cmake --build . --config Release
cd ..

If you want to use the Denoiser class, compiled binaries of Open Image Denoise must be in the correct runtimes folder. For example, on x64 Linux, there should be a libOpenImageDenoise.so in runtimes/linux-x64/native/. See the copy operations in make.ps1 for details. The project works without these binaries in place, but then any attempt to use the Denoiser class will result in a DllNotFound exception at runtime.

When using binaries, especially the official ones, be aware that packaging for .NET requires the RPATH of the shared library to include the same folder that contains the library itself. Otherwise, TBB will not be found. If you don't understand what that means, or how it can be achieved, check out the build script in RenderLibs. (This does not apply to Windows, since the linker there has this behavior by default.)

Build the C# wrapper and run the tests:

dotnet build && dotnet test

To see if the denoiser is linked correctly, you can additionally run

dotnet run --project SimpleImageIO.Integration

These integration tests assume that you have the tev viewer open and listening to the default port on localhost. But you can also comment out the tev-related tests and only run the denoiser ones.

Building the C# wrapper on other platforms

The SimpleImageIO.csproj file needs to copy the correct .dll / .so / .dylib file to the appropriate runtime folder. Currently, the runtime identifiers (RID) and copy instructions are only set for the x86-64 versions of Windows, Linux, and macOS. To run the framework on other architectures, you will need to add them to the .csproj file. You can find the right RID for your platform here: https://docs.microsoft.com/en-us/dotnet/core/rid-catalog.

Note that, currently, Open Image Denoise is included in binary from. The Denoiser class can therefore not be used on platforms other than x86-64 Windows, Linux, or macOS. Attempting to use it on other platforms will cause a DllNotFound exception.

Then, you should be able to follow the steps above and proceed as usual.

Building the Python wrapper

The simplest route is to run the build script

pwsh ./make.ps1

which builds and installs the Python lib with pip, using whatever python executable is currently in the path.

If you need manual control, e.g., specific Python version, here are the required steps:

cd ./FlipViewer
npm install
npm run build
cd ..
cp ./FlipViewer/dist/flipbook.js PyWrapper/simpleimageio/flipbook.js

python -m build
python -m pip install ./dist/simpleimageio-*.whl

The first commands build, bundle, and pack the frontend code. Then, we build the Python package itself and install it via pip. The * must be substituted by the correct version number and runtime identifier.

The tests can be run via:

cd PyTest
python -m unittest