There are a lot of computer graphics papers that haven’t been implemented in any major app, especially in simulation. You could probably pick any SIGGRAPH simulation paper and base a research project around it.
What I’m really interested in is NPR rendering- we’ve got a great line renderer already built-in (Freestyle), and LANPR is probably gonna be in 2.81- but I wonder if there’s research to be done in tone-mapping and building NPR features into Cycles? Cycles is desperately in need of new toon shaders. You might enjoy working with the Mitsuda Renderer if you’re doing research involving rendering, it’s not Blender but it’s open source and designed fro research.
Something that would be really valuable to Blender users is pipeline management/profiling/overview tools. For example, a way of managing the relationships between linked .blend files, tools for measuring performance impact on a per-collection/per-object basis, maybe even research into version controlling 3D binary data (which is kind of hard to do right with traditional VCS, since they’re made for code). Anything involving pipeline integration is a good idea: vdb, alembic, usd, etc.
I’d love to see light-groups for Cycles and especially EEVEE There are a lot of people that want the info panel to automatically generate macros in the same way that MEL does. Texture Painting is a hugely important niche that’s not being filled very nicely by FOSS compared to Substance, Mari, etc. . Implementing UDIM painting might be a good task to work on.
I don’t know how hard you want to work on this; some of these ideas could be a lot more than four months of work. I tried to make a list of idea that either aren’t being worked on or aren’t current targets. And of course, you should scroll through RightClickSelect and see what you like, check the modules on developer.blender.org page and see if any of the roadmaps interest you. You might also look at past GSOC proposals and see if you can revive or improve on them.
If NPR rendering interests you, I can give you some specific ideas about topics to research.
That was an excellent reply, thank you so much for that!
Regarding NPR, it’s not something I’m personally interested in, so it doesn’t motivate me that much, if I’m being frank.
However, all the other suggestions you mentioned are really, really interesting.
I’ve never heard before of the Mitsuba Renderer, but it’s a fascinating beast.
The pipeline tools are extremely valuable, though I am not quite sure whether I’m up for a task like that (maybe it’s the nerves getting to me :D).
The texturing side is one aspect that I never really enjoyed doing on Blender. Substance is light years ahead on that front, so it would be a personal challenge to help raise that bar for FOSS, but one that I would be motivated to do.
Once again, thank you for taking the time to suggest this, I’ll be thinking about this more in the days to come
You might look at Armour Paint, then, if you’re interested in texturing. It’s the only other FOSS app I know of that attempts to fill the niche. It’s kind of a pain to compile but Jayanam has a good tutorial for it.
Rather than necessarily making it about an improvement to Blender, you could still use Blender as the implementation test bed for just about any topic in CS. Something in data structures with an application in Blender, some topic in Rendering where you use Cycles say as the system to implement the idea in (lots of papers out there as someone mentioned), or just about anything you can drive from Python (Machine Learning, or any other Big Data topic) which you could then generate interesting visualizations of using Blender’s Python API, etc.
That makes sense, though I’m a bit unsure whether that would be an interesting thesis from a more practical point of view.
However, there is a possibility of me having the thesis in a company, and the company I’m currently working for is investigating using Blender for server side rendering. Maybe I can join the two of them together
I’m the person who worked on a spectral version of Cycles. I’m more than happy to have a chat with you about my findings and discuss some technical details about it if it is the sort of thing you’re interested in. The engine implementation is one aspect but then there is the other aspect of ‘how do you present a spectral rendering workflow to users in a friendly and effective way?’ Both are interesting and complex problems.
I’ve been seeing some more about Spectral Rendering, and it is a fascinating subject, so fascinating in fact that I already took note on my personal list of possible suggestions!
Hi, I have a branch of 2.79 which was mostly functioning, except for volumes and I think a few light types. Although the visual benefit is very small if you cannot somehow define spectral colours. That’s the biggest issue I ran into, the work in converting the engine is one thing, the harder task is to expose spectral data to the user in an easy to use way.
I found my work, I could either send you a diff file or publish a branch somewhere. Just let me know if you decide to pursue this. I’d love to see it get some more attention and looking back at the work that I did in 1 week (thanks git log) I’d say it is certainly achievable.
Well, I would love something that I could finish from beginning to end! I am moderately freaking out if I’m being fully honest If I do end up pursuing this, I’ll let you know so I can look at your work with more attention. Even if I decide I won’t do this, I’ll still give you a heads up so as to not leave you hanging!
If someone else has any more suggestions, I’d love to hear what you think!