Do you connect meshes or not?
vertex
Sorry if i worded the title incorrectly, but lets see if i can word it better here. What i'm trying to say is that if you were making 1 object and needed to use 2 different polygons would you try to connect their meshes ? Or just combine them to be 1 object? I want to know because i don't know if that would screw up in animation or not. Thanks for any replies!
Replies
Preferably, you don't want to have floating/intersecting geometry as it leads to z-fighting. I've also heard it can affect hardware performance, as more individual meshes = more draw calls.
However, I do know some tricks were it is necessary to separate a mesh. For example, on a gun model, you would model the trigger but you would select the polygons that make up the trigger and "disconnect" it from the rest of the model. That makes it easier to animate a gun firing, so I've heard.
Well let me use this as an example now, i'm making this lamp type model and i want the ends to flare out as such, but i didn't really see a way to make it with one piece without disfiguring the top piece, so would i try to connect these 2 or leave them as is?
Jordan - "more individual meshes = more draw calls." this is true, but it isn't true when we are talking about sub meshes or lets call them elements.
"Preferably, you don't want to have floating/intersecting geometry as it leads to z-fighting." This also isn't completely true, then how you'd make decals? Also intersecting geo can be also ok. Think of when you are putting rock meshes together to get rock formation. You can get Z issues, but there are many ways to handle it. Also Z precision can be set.
On a gun model you'd split a lot of parts so they can be animated. Still the best rule to follow is, to split the mesh based on how the real life version would be split. Its really rare to make things from one continuous mesh. Only if the real life version would be from one continuous piece too.
http://www.polycount.com/forum/showthread.php?t=117017
https://docs.unrealengine.com/latest/INT/Engine/Content/Types/StaticMeshes/LightmapUnwrapping/index.html
+ texture space saved
- probably more verts
- more work when starting from high poly source
I don't think you get better bakes from continuous meshes. I intersect geometry all the time and get fantastic bakes. The only time I worry about having a continuous mesh is for the purpose of deformation. Weight painting is much simpler on a continuous mesh.
Here is a thread related to overdraw:
http://www.polycount.com/forum/showthread.php?t=89154
And you can read this in it:
"Overdraw:
It discribes the number of polygons rendered ontop of a single" pixel."
Also here is the related wiki page:
http://wiki.polycount.com/wiki/Overdraw
So as I understand there can be 2 cases of overdraw:
1. transparency, and things behind it.
2. When you view a model from very far and there are multiple triangles on a single pixel. So then in this case you get it very often.
So I think overdraw is not really related to baking and watertight geo.
As for baking, and the relation of the topic. There are some cases when you do want to connect meshes, especially when you use lower poly count limits. It can also give nicer edges at the meeting points in some cases, but sometimes you want it differently, especially when you can use higher poly count. I used both methods, and both has its positive and negative side.
This is how I see and do it.
Also sorry if my questions are a little hard to understand it's hard to put into words what i'm trying to say.
Raycast errors on cage intersects, for me.
- left low poly is merged, right low poly is intersecting
- 20 vertices vs. 16 (left/right)
- high poly is merged with beveled edge between the two primitives
- didn't edit cage, just a simple push outwards
Thanks for taking the time, but this is just one case, and there can other ones, and then it depends.
@YellowYeti - car would definitely be from multiple parts. Also its rare to bake a car's body
In the case of your lamp, it depends. Are you planning to make highpoly? If yes, then how it would look? How much parts you would use.
1. if you don't explode the separate parts you get raycast errors at the intersection
2. if you do explode the separate parts your AO lacks contact shadows between exploded parts, leading to
2a. the need for both "combined" and "exploded" AO bakes
3. you waste UV space on unseen geo; downsampling can grab bad data
i merge nearly everything that doesn't have to move
To make it clear, I also combine some things, but not everything.
Also here is an another thing, which wasn't covered yet. When you combine, you get more verts, as its discussed. But usually you need hard edges at those transitions so it means even more extra verts.
read past point one in the explanation that you asked for
doesn't add much in practice. might be worrisome for mobile / facebook spec
But do you even need ao at all if you're going for pbr? I was under the impression it was more or less obsolete. Sorry if I'm saying something stupid.
AO is still useful for placing dirt rust or other such things in occluded areas, even though AO itself is no longer multiplied over diffuse
Another occasionally overlooked plus when considering merging lowpoly mesh items is that you also save UV space. There's also some good info in this thread: http://www.polycount.com/forum/showthread.php?t=117017
There are plenty of exceptions and positives as have been noted in this thread.
I personally make most of my low polys that are baked from high watertight.
Because on example b, you need to have the whole box uved, and a big chunk of it is being hidden because it's intersecting the other box.
example A is more sensible because even tho you have more splits, your're not wasting pixels with things that won't get visible.
does that make sense?
Here is my example:
Lets say you have a box, and there are some modeled screws on the lp of it too. Would you still connect them to the box? I obviously wouldn't.
and also, i prefer method A, i was just explaying how method b is more prone to wasted pixels.
I still think this "to connect or not to connect" thing depends on the case. It isn't always the best option in my opinion.
Taking the above example at face value; I probably wouldn't merge the screws because it'd be easier while working to have 3 of them an instance of 1, so they'd share UV's. This could still be done when merged but the texture space saved isn't significant so I'd spend the time elsewhere.