This project contains a simple C++ library to convert hdri (equirectangular) environment maps to cubemaps, an example of which is shown below, for an hdri obtained from HDRIHaven:
To determine the 'color' of a pixel in one the cubemap's faces:
- The direction in 3D world space corresponding to the pixel is determined.
- The corresponding azimuth and elevation are calculated
- The environment map is sampled at the pixel corresponding to the calculated azimuth and elevation (for the equirectangular environment map the pixel distance along x and y are proportional to azimuth and elevation respectively)
Currently, stb image is used for image loading and writing, providing support for many image formats.
OpenCL support is provided for calculating the cubemap's faces. Do note that there is no big performance gain compared to the cpu implementation, since io operations take up most of the time.
The only dependency of HdriToCubemap is stb image. Clone the repository for an example implementation using:
git clone https://github.com/ivarout/HdriToCubemap.git
cd HdriToCubemap
git submodule update --init --recursive
When building through the CMake Gui, the HDRITOCUBEMAP_USE_OPENCL
option can be checked if the user wishes to use the OpenCL implementation.
Note that C++17 is required (due to usage of std::clamp).
Note that the template class HdriToCubemap
supports float
and unsigned char
types, the first storing an hdr image, and the latter an 8-bit-per-channel ldr image. If the provided image is an hdr image, while the provided template type is <unsigned char>
, stb image will perform the conversion from hdr to ldr, and vice versa.
When constructing an instance of the HdriToCubemap
class, the user has to provide a resolution to be used for the cubemap faces, and optionally a boolean which sets the filter type, which defaults to true
, resulting in bilinear interpolation. When set to false
, the nearest pixel in the environment map will be used.
// load hdr or ldr image
int cubemapResolution = 1024;
bool linearFilter = true;
HdriToCubemap<float> hdriToCube_hdr("path_to_hdri_file", cubemapResolution, linearFilter);
HdriToCubemap<unsigned char> hdriToCube_ldr("path_to_hdri_file", cubemapResolution, linearFilter);
// use cubemap image directly
unsigned char* front = hdriToCube_ldr.getFront();
if (hdriToCube_ldr.getNumChannels() == 4)
{
glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_X, 0, GL_RGBA8, hdriToCube.getCubemapResolution(), hdriToCube.getCubemapResolution(), 0, GL_RGBA, GL_UNSIGNED_BYTE, front);
}
// or output cube images to file (6 images will be output: "front.png", "back.png", "left.png", "right.png", "up.png", and "down.png")
hdriToCube_ldr.writeCubemap("output_directory");
To use OpenCL (v1.2) for calculating cubemap faces, _USE_OPENCL
has to be defined. See CMakeLists.txt for reference.
Any comments or suggestions related to code improvements and/or bugs are appreciated.