eradiate-mitsuba 0.5.0

Eradiate radiometric kernel, based on the Mitsuba 3 renderer

Eradiate Radiometric Kernel — Mitsuba 3

Introduction

This is the radiometric kernel of the Eradiate radiative transfer model, based on Mitsuba 3, the rendering system developed at EPFL in Switzerland. This modified version builds on Mitsuba 3's feature set to provide a modern technical foundation for the implementation of high-performance Monte Carlo ray tracing techniques to solve atmospheric radiative transfer problems.

How does this project relate to Mitsuba 3?

The Eradiate kernel is a fork of Mitsuba 3. It inherits many of its distinctive features, such as retargetability (the ability to apply compile-time transformations to the C++ codebase to change fundamental aspects of the simulation such as spectral representation or polarization), and is powered by the same just-in-time compiler Dr.Jit.

However, the Eradiate kernel is more specialized than upstream Mitsuba 3:

• Monochromatic computation: Currently, Eradiate hands the management of the spectral dimension to pre- and post-processing components external to the radiometric engine. Consequently, all radiometric computations are monochromatic.

• Double precision: Earth observation problems involve a wide range of length scales: objects of a size of ~1 cm, such as leaves, coexist with objects of a size of ~10.000 km, such as planets. In such conditions, performing ray tracing safely is challenging. For safety, Eradiate currently relies only on double-precision variants, which has two consequences: (1) the high-performance BVH supplied by Embree cannot be used (it supports only single-precision), and we must fall back to the built-in kd-tree; (2) using JIT-compiled variants targeting the GPU is not yet possible. Addressing these limitations is planned.

• Scalar variants only: For now, only scalar variants are used. Supporting JIT compilation is planned.

About

The Eradiate kernel is maintained by Rayference.

Mitsuba was created by Wenzel Jakob, and the Eradiate team thanks all Mitsuba contributors, in particular EPFL's Realistic Graphics Lab, for their work and support.