Agreed with the "feedback guy". It's really hard to understand the concepts and very few guides. (Especially Shadow backing. Had to test out different light modes for 4-5 hours just to understand 3 modes)
Very helpful talk. I just gave some of the tips a try and managed to get baking times down from 2 minutes to 2 seconds for a simple test scene. Three bits that were missing for me: a) Since I can only set the lightmap scale for the entire object, does that mean I should split a big static mesh (let's say I made an entire room as one mesh) into several chunks to get this done? b) Some examples for different sample counts in the lightmap settings would have been nice to give a rough idea what settings might be sensible for different scenarios. c) I was really hoping for an ETA on the GPU lightmapper. It seems to have disappeared from the roadmap since GDC and is now listed under "Development" without an ETA.
a: Yes. Generally, you'll have more win and less struggle if you import scenes into Unity as a tree of meshes (a GameObject hierarchy, each with their own Mesh & MeshRenderer) and then rely on Unity's static batching to reduce draw calls in the resulting game. Not only does this give you much more power in Unity to texture/material/shade different parts of the scene differently, to potentially animate or disable parts of the scene later (in game dev, tweaking and tuning can be necessary at any time, and immediate feedback is super important), and to optimize physics meshes/primitives to just what's necessary, it also allows finer-turned control of lightmapping. b: Agreed. c: It's out now (a year later), so I'm sure you're aware, the wait is over.
Ah ok, sorry for being sharp. Yeah vxgi is awesome but it involves re-rendering the scene pretty inefficiently into a whole bunch of 3D voxel blocks, then generating a bunch of 3D mipmaps, then reading tons of those 3D voxel blocks to compute lighting. This is so expensive it's usually done at a lower resolution and upscaled from there, which only really works on deferred rendering (so no gi on transparency). It also by default has lots of light bleeding through walls. which is why it mostly get's shown in brightly lit outdoor scenes, or single room interiors. There are solutions to many of these problems but they all take a lot of extra complexity and performance tradeoffs. We're a ways off from true realtime gi sadly =
Finally someone explains this lightmapper. Thanks for uploading. This helps a lot.
Agreed with the "feedback guy". It's really hard to understand the concepts and very few guides. (Especially Shadow backing. Had to test out different light modes for 4-5 hours just to understand 3 modes)
so
1. You get proper specular only with Baked Indirect?
2. Why lightmaps has overlapping UVs even when I enable "Generate Lightmap UVs"?
I 'm baking lightmaps while watching this :o)
we all are!
Hahaha, here too! I was having problems and this solve all my problems
ETA 389:56:25
31:33 GPU baking
35:16 future
Lots of useful informations. Great work.
awesome talk
GPU light mapping is still in eternal beta unfortunately
Very helpful talk. I just gave some of the tips a try and managed to get baking times down from 2 minutes to 2 seconds for a simple test scene.
Three bits that were missing for me:
a) Since I can only set the lightmap scale for the entire object, does that mean I should split a big static mesh (let's say I made an entire room as one mesh) into several chunks to get this done?
b) Some examples for different sample counts in the lightmap settings would have been nice to give a rough idea what settings might be sensible for different scenarios.
c) I was really hoping for an ETA on the GPU lightmapper. It seems to have disappeared from the roadmap since GDC and is now listed under "Development" without an ETA.
a: Yes. Generally, you'll have more win and less struggle if you import scenes into Unity as a tree of meshes (a GameObject hierarchy, each with their own Mesh & MeshRenderer) and then rely on Unity's static batching to reduce draw calls in the resulting game. Not only does this give you much more power in Unity to texture/material/shade different parts of the scene differently, to potentially animate or disable parts of the scene later (in game dev, tweaking and tuning can be necessary at any time, and immediate feedback is super important), and to optimize physics meshes/primitives to just what's necessary, it also allows finer-turned control of lightmapping.
b: Agreed.
c: It's out now (a year later), so I'm sure you're aware, the wait is over.
Lightmapping eyeballs: "Be not afraid"
Its more like a Turorial... a really great one :-D
(But sadly, its weirdly stuttering)
Sorry about the stuttering! Unfortunately, we can't do anything about it. Happy you've enjoyed the video anyway though!
Could you create voxel-Based Global Illumination so that it wouldn't need to precompute GI ?
You do realize how expensive VXGI is right? It barely runs on the highest end of PCs for real scenes. I love it but it's not currently practical.
I didn't know that. I only saw it in Cryengine without the need to bake light.
Ah ok, sorry for being sharp. Yeah vxgi is awesome but it involves re-rendering the scene pretty inefficiently into a whole bunch of 3D voxel blocks, then generating a bunch of 3D mipmaps, then reading tons of those 3D voxel blocks to compute lighting. This is so expensive it's usually done at a lower resolution and upscaled from there, which only really works on deferred rendering (so no gi on transparency). It also by default has lots of light bleeding through walls. which is why it mostly get's shown in brightly lit outdoor scenes, or single room interiors. There are solutions to many of these problems but they all take a lot of extra complexity and performance tradeoffs. We're a ways off from true realtime gi sadly =
Thx for clarification.
voxels not good for final product in current hardware, it next-next gen
Still don’t understand about texel LoL