glTF IBL Sampler

Introduction

This project pre-filters an environment High Dynamic Range (HDR) panorama image and stores diffuse and specular parts in cube maps respectively. Considering different material characteristics, the specular part consists of several mip-map levels corresponding to different roughness values of the modeled material. The final basis compression into a KTX2 file ensures small file sizes while maintaining reasonable image quality.

Algorithm references:

• Real Shading in Unreal Engine 4
• GPU-Based Importance Sampling
• Runtime Environment Map Filtering for Image Based Lighting
• Filtering shader

Workflow

The individual transformations can be executed separately in the library, enabling the use of a cube map as input in future releases.

• Environment HDR image → Cube Map
• Cube Map → Filtered Cube Map
• Filtered Cube Map → KTX2 Output File

This separation of shader passes ensures both, flexibility and extensibility. The pre-filtering will use the algorithms noted in the glTF Sample Environments repository.

Building

The project provides a cmake file to generate respective build files.

Third Party Requirements:

• Vulkan SDK
• Glslang (included in the Vulkan SDK)
• STB image library (git submodule, no need to install)
• KTX-Software (you might need to manually install KTX-Software with the pull request that fixes cmake find_package)

CMake option IBLSAMPLER_EXPORT_SHADERS can be used to automatically copy the shader folder to the executable folder when generating the project files. By default, shaders will be loaded from their source location in lib/shaders.

The glTF-IBL-Sampler consists of two projects: lib (shared library) and cli (executable).

Usage

The CLI takes an environment HDR image as input. The filtered specular and diffuse cube maps can be stored as KTX1 or KTX2 (with basis compression).

• -inputPath: path to panorama image (default) or cube map (if inputIsCubeMap flag ist set)
• -outCubeMap: output path for filtered cube map (default=outputCubeMap.ktx2)
• -outLUT: output path for BRDF LUT (default=outputLUT.png)
• -distribution: NDF to sample (Lambertian, GGX, Charlie)
• -sampleCount: number of samples used for filtering (default = 1024)
• -mipLevelCount: number of mip levels of specular cube map. If omitted, an optimal mipmap level is chosen, based on the input panorama's resolution.
• -cubeMapResolution: resolution of output cube map. If omitted, an optimal resolution is chosen based on the input panorama's resolution.
• -targetFormat: specify output texture format (R8G8B8A8_UNORM, R16G16B16A16_SFLOAT, R32G32B32A32_SFLOAT)
• -lodBias: level of detail bias applied to filtering (default = 0)

Example

.\cli.exe -inputPath ..\cubemap_in.hdr -outCubeMap ..\..\specular_out.ktx2 -distribution GGX -sampleCount 1024 -targetFormat R16G16B16A16_SFLOAT
.\cli.exe -inputPath ..\cubemap_in.hdr -outCubeMap ..\diffuse_out.ktx2 -distribution Lambertian -sampleCount 1024 -targetFormat R16G16B16A16_SFLOAT