Some News

03 Dec 2020

2020

This year was a tough one for most of us. Not just because of COVID-19. Let me start with a personal email I did send to 3D-PRO, a private, carefully curated mailing list, which is around for over 20 years:

Subject: Time to say goodbye?

Hi,

The last couple of weeks I did not really announce that I was affected
by the recent Technicolor layoffs, because (thanks to German Labor
Law) I did have some time to think about my past and how I imagine my
remaining work life (I'm 53 now and counting).

In the past I have done NURBS libraries (more than 25 years ago),
international projects with supercomputing and CFD/NURBS (for
Fraunhofer research), VR/AR with expensive computers (SGI Onyx 2 - 24
years ago), I worked on Blender (20 years ago) before it became open
source (we had already a Python API back then), helped building some
infrastructure for the International Net Management Center (INMC) of
Deutsche Telekom in Frankfurt, and ended up in film and commercials
(MPC 2001, Mill Film 2001, The Mill 10 years and 20 month in total, DD
2005, mi/Nvidia 2006). I have done my share of work related
travel/residence (Germany, Netherlands, UK, USA) and are not planning
to continue like that (my dad, an artist, died and we inherited
hundreds of his paintings, sculptures, photos, film) and I became
engaged. So, no reason for me to leave Berlin (again). I do have
experience working from home (not just since Covid 19), on and off
since 2000.

I have written plugins/addons for 3DS (DOS and Max), Blender,
Softimage XSI, Maya, Houdini, have written shaders for RenderMan,
mental ray, MetaSL, Cg/GLSL/HLSL, Arnold, and finally my own renderer
in Rust:

https://www.rs-pbrt.org/about

https://github.com/wahn/rs_pbrt (100.4K lines of code)

In a perfect world I would work with Rust as a programming language
(but of course there are more mainstream languages like C/C++, Python,
etc. I have to consider and used in the past to get my jobs done), and
Blender and/or Houdini as DCC of choice (I'm not a TD though). I do
see a bright future for USD:

https://www.janwalter.org/Publications/DP2005_049-053_USDView_HiRes.pdf

What I don't see is a bright future in this industry, but I could be
wrong.

So, I'm not saying goodbye (yet). I will keep lurking around for
another month or to the end of this year. If you would like to respond
to this email (personally) or have a job offer (working from home in
or Berlin), please write me an email (off the list) or contact me via
LinkedIn:

https://www.linkedin.com/in/janwalter

Have fun and stay healthy,

Jan

P.S.: Sure, I will not be able to ignore C++ for the rest of my work
life, and there is C++03, C++11, C++14, C++17, C++20, and C++23 on the
horizon, but here is a long read (and why I prefer Rust):

https://dl.acm.org/doi/10.1145/3386320 (read the PDF)

So the year is coming to an end. I haven't signed a new contract yet, but I'm positive that this will happend within 2020. So beside giving you a little history of my working life, I mention a couple of things in the email above. For example the article from Bjarne Stroustrup about Thriving in a crowded and changing world: C++ 2006–2020 might be interesting not just for C++ fans (or haters), but how things develop over time and how to react (or not react) to changes in HW etc. Lessons learned ... for the evolution of programming languages (in general) ...

Which brings me to a newer language, Rust, I started to like a lot, and used to write my own renderer (to learn the language, and see how easy or difficult it is to translate an existing C++ code base to a new language).

RS-PBRT

There is a whole web page I created about rs-pbrt and I'm not going to repeat the information you will find there, but basically it's a feature complete Rust-based implementation of the 3rd edition of the PBRT book (the book ships with C++ code).

Landscape view 3.

There is a fourth edition of the same book on the horizon and the C++ source code is already available.

v3

But let's first talk about the third edition and it's Rust counterpart. In contrast to the C++ version I ship a couple of executables (and example scenes) with the Rust-based source code, which can render images:

$ ./target/release/rs_pbrt -h
pbrt 0.8.3
Parse a PBRT scene file (extension .pbrt) and render it

USAGE:
    rs_pbrt [OPTIONS] <path>

FLAGS:
    -h, --help       Prints help information
    -V, --version    Prints version information

OPTIONS:
    -t, --nthreads <nthreads>    use specified number of threads for rendering [default: 0]

ARGS:
    <path>    The path to the file to read

To render for example the Cornell Box (.pbrt is a file format defined in the book for v3 and will most likely change a bit for v4):

$ rs_pbrt assets/scenes/cornell_box.pbrt
pbrt version 0.8.3 (v0.8.3-4-g78bdee9) [Detected 28 cores]
Copyright (c) 2016-2020 Jan Douglas Bert Walter.
Rust code based on C++ code by Matt Pharr, Greg Humphreys, and Wenzel Jakob.
Film "image"
  "integer xresolution" [500]
  "integer yresolution" [500]
Accelerator "bvh"
Sampler "halton"
  "integer pixelsamples" [128]
Integrator "path"
  "string lightsamplestrategy" ["spatial"]
Rendering with 28 thread(s) ...
1024 / 1024 [===============================================] 100.00 % 126.66/s
Writing image "pbrt.png" with bounds 
Bounds2i { p_min: Point2i { x: 0, y: 0 }, p_max: Point2i { x: 500, y: 500 } }

More v3 scenes can be found on the pbrt.org web site or on GitLab (tested versions for the Rust-based renderer). So, please, read the Getting Started page, compile the source code, and run some example scenes to see for yourself.

What most people don't know is that I wrote a little example program for Arnold's scene description (.ass extention) as well:

$ ./target/release/examples/parse_ass_file assets/ass/cornell_box.ass 
parse_ass_file version 0.8.3 (v0.8.3-4-g78bdee9) [Detected 28 cores]
FILE = "assets/ass/cornell_box.ass"
search_directory is /home/jan/git/github/rs_pbrt/assets/ass
5174 bytes read
options {
 xres 500 
 yres 500 
 camera "CACamera" 
 GI_total_depth 11 }
standard_surface { name MAdefault }
standard_surface { name MAbox_Material 
 base_color 0.5 0.5 0.5 
 specular_color 1 1 1 }
standard_surface { name MAcbox_Material 
 base_color 0.4 0.4 0.4 
 specular_color 1 1 1 }
standard_surface { name MAcbox_red 
 base_color 0.5 0 0 
 specular_color 1 1 1 }
standard_surface { name MAcbox_green 
 base_color 0 0.5 0 
 specular_color 1 1 1 }
render_camera = "CACamera" 
fov = 39.14625 
filter_name = "gaussian"
filter_width = 2.0
max_depth = 11
samples_per_pixel = 16
number of lights = 2
number of primitives = 36
Rendering with 28 thread(s) ...
1024 / 1024 [===============================================] 100.00 % 711.28/s
Writing image "pbrt.png" with bounds
Bounds2i { p_min: Point2i { x: 0, y: 0 }, p_max: Point2i { x: 500, y: 500 } }

It's not really meant to render an arbitrary Arnold scene, but it was a proof of concept, that a parser for .ass files could be written and the API of the Rust-based renderer could be used to render such a scene (using some heuristics to translate shaders into a PBRT material).

Last, but not least, Blender scenes (for now I focused on version 2.79 for examples provided by myself) can be rendered like this:

$ parse_blend_file assets/blend/suzanne_integrator_test_2_79.blend 
parse_blend_file version 0.8.3 (v0.8.3-4-g78bdee9) [Detected 28 cores]
BLENDER-v279
857932 bytes read
number of lights = 2
number of primitives = 5426
500x500 [100%] = 500x500
integrator = "path" [(Unidirectional) Path Tracing]
  pixelsamples = 1
  max_depth = 5
Rendering with 28 thread(s) ...
1024 / 1024 [==============================================] 100.00 % 6248.05/s
Writing image "pbrt.png" with bounds 
Bounds2i { p_min: Point2i { x: 0, y: 0 }, p_max: Point2i { x: 500, y: 500 } }

This executable is useful to render the same scene with different settings (and render algorithms) and/or cameras without changing the scene. See the output of the help message:

$ parse_blend_file -h
pbrt 0.8.3
Parse a Blender scene file and render it

USAGE:
    parse_blend_file [OPTIONS] <path>

FLAGS:
    -h, --help       Prints help information
    -V, --version    Prints version information

OPTIONS:
        --bootstrap_samples <bootstrap-samples>
          bootstrap samples [MLT] [default: 100000]
    -c, --camera_name <camera-name>
          camera name
        --chains <chains>
          number of Markov chains [MLT] [default: 1000]
    -i, --integrator <integrator>
          ao, directlighting, whitted, path, bdpt, mlt, sppm, volpath
    -l, --light_scale <light-scale>
          global light scaling [default: 1.0]
    -m, --max_depth <max-depth>
          max length of a light-carrying path [default: 5]
        --mutations_per_pixel <mutations-per-pixel>
          number of path mutations [MLT] [default: 100]
    -s, --samples <samples>
          pixel samples [default: 1]
        --sigma <sigma>
          perturbation deviation [MLT] [default: 0.01]
        --step_probability <step-probability>
          prob of discarding path [MLT] [default: 0.3]
        --write_frequency <write-frequency>
          frequency to write image [SPPM] [default: 1]

ARGS:
    <path>    The path to the file to read

Scenes which were tested to render with rs-pbrt can be found in the GitLab repository as well (in the blend folder). To find out which cameras are defined etc. in a .blend file you can use blend_info which I provide the source code for on codeberg.org. Here some examples:

# find out which cameras are defined
$ blend_info conference_room_v2_79.blend | grep '"CA'
  "CAcurrent_cam" has 2 data blocks
  "CAdoor1_cam" has 2 data blocks
  "CAdoor2y_cam" has 2 data blocks
  "CAshaft_cam" has 2 data blocks
  "CAxY_cam" has 2 data blocks
# use one of the cameras to render
$ parse_blend_file -c current_cam -i path -s 128 conference_room_v2_79.blend
...
integrator = "path" [(Unidirectional) Path Tracing]
...
$ parse_blend_file -c door1_cam -i bdpt -s 64 conference_room_v2_79.blend
...
integrator = "bdpt" [Bidirectional Path Tracing (BDPT)]
...
$ parse_blend_file -c door2y_cam -i mlt -s 1 conference_room_v2_79.blend
...
integrator = "mlt" [Metropolis Light Transport (MLT)]
  pixelsamples = 1
  max_depth = 5
  bootstrap_samples = 100000
  chains = 1000
  mutations_per_pixel = 100
  step_probability = 0.3
  sigma = 0.01
...

If you find bugs, please open an issue or use the Discord server to chat.

v4

As I said before, there is a fourth edition of the same book on the horizon and the C++ source code is already available. You can find a video with Matt Pharr (one of the authors of the book) about the pbrt-v4 code walkthrough. There are also slides and another video about Porting PBRT to the GPU While Preserving its Soul. I'm not sure if I will have time to convert the new C++ code to Rust, but there are a couple of things I would like to do for the Rust-based version. Similar to the C++ version I would like to be able to watch the renderer while the image is created. The C++ version connects to tev, a EXR Viewer (can be seen in one of the videos). The Rust code can be compiled with support for OpenEXR already, but maybe there is an even simpler way to show the progress while rendering? Last, but not least, it would be interesting to see how the GPU code can be done in Rust, similar to those changes in the C++ version.

Which brings me to my last topic (for this year).

Mitsuba 2

The second version of Mitsuba (by Jakob Wenzel, another author of the book mentioned above) is available (the code is on GitHub). There is a video and a paper available and Enoki is another piece of code, which I would like to study and discuss with others.

Job

But first I need to resolve my job situation ... Stay tuned ...

Update

I signed a contract with Parakar Germany GmbH to be able to work remotely for Paramex Limited in the UK. Starting date is Monday, 14th of December 2020.