RenderJay.jl

RenderJay is a Julia-based path tracer that is intended for operation on workstations and (large) compute clusters, with applications in ecology and remote sensing where scenes are characterized by very high polygon counts, lots of detail, and light spectra may be broken down into hundreds of wavebands (imaging spectroscopy), but applications for this package are not so much sought in computer graphics or architecture, although you could. RenderJay uses no bump maps or textures; everything is down to geometric shaped (i.e., spheres, cylinders, cones, disks and triangle meshes) and bidirectional reflectance (transmittance) distribution functions (BRDFs, BTDFs). Being written in Julia, it can profit from Julia's distributed computing capabilities, enabling processing of large workloads across multiple machines.

RenderJay can be installed as follows from the Julia REPL:

using Pkg
Pkg.add("RenderJay")

At the bottom of this page you will find an example code for rendering. It will produce a top-down view of four Cornell boxes in different colours, floating above a gray flat surface. Already produced renderings can be found in the img/ folder and an overview of the contents of the scene files that are used for rendering can be found in the notes/ folder. A small test, demonstrating basic capacity, can be found in the test/ folder.

Plant models

A good place to look for plant models is online at various 3D model warehouses. The Wavefront OBJ file format is very close to what RenderJay uses and a simple script in your favorite language can be used to produce the separate vertex and mesh connectivity files. To see how they should look, check out the assets you find under the test/ folder in this repository. Don't worry about the BVH file (it stands for Bounding Volume Hierarchy and it is for spatial indexing). These are produced the first time you use a new asset. If you ever make changes to an existing asset, do remove the BVH file that is associated with that asset. This will automatically create a new BVH based on the changes you applied to that asset. Besides browsing any 3D warehouses, you can also try out e.g. Arbaro - a tree generator for PovRay that was developed by Weber and Penn 1995. It can be found here: http://arbaro.sourceforge.net/

Performance indication

Rendering a 512x512px image of the below Cornell boxes scene took 00:34:04 (HH:MM:SS) on the Dell T7910 (DUAL E5-2630V3) using 30 workers and it took 3:39:20 on a Lenovo Edge 15 laptop (i7-4510U) with only 3 workers. For some very simple scenes with only a few cylinders on a flat surface (not shown) up to a million rays per second were processed using 30 workers.

Applications

RenderJay can be used to study how different plant compositions or canopy structures affect spectral mixing at a given spatial resolution or viewing geometry. It can be used to derive hemispherical photographs, spectral signatures, and (bi)directional surface reflectance, among other quantities. For a demonstration, this study in the Greater Cape Floristic Region of South Africa uses the RenderJay model to assess the effects of spectral mixing across a range of pixel sizes and species compositions (https://doi.org/10.1016/j.rse.2021.112405).

RAMI-5

As part of the RAdiative transfer Model Intercomparison phase 5 (https://rami-benchmark.jrc.ec.europa.eu/_www/phase_descr.php?strPhase=RAMI5), we have prepared RenderJay scene files for the various test cases involved, including abstract scenes, fully parameterized canopies, and (semi-)empirical forest canopies. Some of the illustrations below show raw renderings where no white-balancing has been applied yet, which causes the Wytham Woods scene to appear rather greenish. The irradiance conditions for these example renderings were set to a 10% diffuse sky with a watery sun.

Here is an image showing the downwelling irradiance map that was used for these renderings, with bearing along the horizontal axis and zenith along the vertical:

After applying white-balancing on the snow of the Wytham Woods scene, for example, we get a realistically looking and pleasing rendering:

(feel free to contact Martin van Leeuwen to obtain RenderJay scene file for the RAMI5 exercises)

And here are two overhead views of the Jarvsilja Pinestand (summer) and Ofenpass (winter) scenes. You can clearly see that there is a discontinuity along the bearing of the irradiance map -- which, for the purpose of demonstration, has the benefit that it shows how the irradiance map drapes around the visible sky.

By comparing the amount of radiance scattered into a particular hemispherical direction and comparing this with the amount of radiance that is scattered into that direction by a Lambertian reference target, we can derive the Bidirectional Reflectance Factor (BRF). With RenderJay you can use cyclic boundaries, as is needed for the RAMI exercises, so that any radiation that leaves the scene in a lateral direction, re-enters back in from the opposite side and rays are traced until they leave the scene from the top plane or ceiling that is spanned just above the highest scene element. In the polar plot below, the green dot indicates the position of the sun. (The white "+" signs indicate the 0, 45, and 90 degree marks.)

Example code

Below is an example that renders a set of four Cornell boxes with different colours floating just above a flat surface. If you are on a laptop or you have few logical cores, please mind the addprocs() line and set the number to something comfortable, e.g., the number of logical cores that are available.

using Distributed
addprocs(3)
@everywhere using RenderJay
using CSV, LightXML, DataFrames, DelimitedFiles, LinearAlgebra, ProgressMeter, SharedArrays, ImageMagick, Images

pth = pathof(RenderJay)
src_folder, _ = splitdir(pth)
root_folder = src_folder[1:end-3]
scene_fn = joinpath(root_folder, "test/testplot_little_cornellboxes.xml")
assets, geometries, palettes, geometry_bvhs, scene_bvh, skymap, camera = read_scene(scene_fn);
coords = create_coords(camera);
img = SharedArray{Float64, 3}(camera.nBands, camera.xResolution, camera.yResolution);

@time @sync @distributed for coord in coords[1:nprocs()]
    img[:, coord.x, coord.y] = render_pixel(coord, assets, geometries, palettes, geometry_bvhs, scene_bvh, skymap, camera)
end

npixels = length(coords)
blocksize = camera.xResolution
@showprogress for i=1:blocksize:npixels
    e = min(i+blocksize, npixels)
    @sync @distributed for coord in coords[i:e]
        img[:, coord.x, coord.y] = render_pixel(coord, assets, geometries, palettes, geometry_bvhs, scene_bvh, skymap, camera)
    end
end

You can save the rendering to disk as follows:

using ImageMagick, Images
img3 = reshape(hcat(img...), camera.nBands, camera.xResolution, camera.yResolution)
mn, mx = minimum(img3), maximum(img3)
imscl = img3 ./ (mx/1.0)
imscl .*= 10.0
msk = imscl .> 1.0
imscl[msk] .= 1.0
im = colorview(RGB, imscl)
save("/tmp/test_little_cornell_boxes.tif", im')
a = 1.5
b = 10.0
_, imwb = weibull(imscl, a, b)
im2 = colorview(RGB, imwb)
save("/tmp/test_little_cornell_boxes_wb.tif", im2')

Et voila! You should be seeing something like this.

Testing

Basic test functionality is included in the test/ folder and may be run from the REPL as follows:

julia> using Pkg
julia> ]
pkg> test RenderJay