Standard ways to create xml scenes for research purposes

Question

I am a newbie in computer graphics and just started working on physically based rendering engines such as Mitsuba.

I found it difficult to create my own scenes, that are xml files. To create one, there are three ways as I know so far:

1. Build a scene in Blender and export it to a xml file using prasers, e.g. https://www.mitsuba-renderer.org/plugins.html.

2. Download free xml scenes, e.g. from https://benedikt-bitterli.me/resources/.

3. Hand-write an xml scene with Python from scratch, e.g. this example http://www.djmannion.net/psych_programming/vision/mitsuba/mitsuba.html.

However, I am suffering from following issues:

• It takes long time for me to create a scene via 1 and 3, yet not customizable via 2.

• I do not know where to get free meshes (and license issues?)

Questions

• Are there any better ways to create an xml scene?

• What is the canonical way to create delicate scenes, e.g. the ones in SIGGRAPH papers?

Answer 1

(In case someone encounters similar issue)

I end-up in Way 1, and it seems the standard way in academia.

Once I am familar with Blender, the scene creation becomes less challenging; still, it takes some time to create a simple scene. Fortunately for academic papers the number of scenes are a few (< 10).

Unfortunately the old Mitsuba parser does not work for the latest Blender version. So I wrote a custom 2.8x Blender add-on for Mitsuba, as well as creating new, custom material nodes for my own renderer.