The pros and cons of contiguous and non-contiguous mesh baking?

DirtyBlueFL

I made a quick example to highlight the topic I'd like to learn more about.
From left to right: High poly with 2 meshes, low poly A with 2 meshes, low poly B with 1 mesh.
Note that low poly A's larger mesh has been rotated so that the planes are not parallel with the planes of the sub-mesh cap.

What I think I understand:
-Low poly A requires 2 bake groups for good normals. The AO bake will be a little off around the intersecting meshes, especially due to the non-co-planar faces, but would be negligible. Lower tri-count than low poly B. Less trouble with seams when using organic / patterned textures.

-Low poly B requires 1 bake group for good normals, and AO looks more accurate. Non-co-planar faces aren't a problem, as the mesh is contiguous, and captures all the information. Higher tri-count than low poly A. More use of hard edges and seams, which will require use of tri-planar projection in substance painter, and/or more finesse when dealing with patterns in textures. 

What I would like to know more about:
When using auto-LOD generation in UE4, is it "better" to use non-contiguous meshes? I imagine that the engine does a better job reducing sub-meshes, as opposed to grabbing verts with seams, and welding them to their nearby partners.

for the sake of portfolio, what's more impressive; using Low poly A, and working to get the unrealistic inset meshes to look great? Or using low poly B, hiding seams in plain sight, and having a much more visually pleasing bake?

My actual project, that I can't show you here, has many fine, ornate, details, that break up the objects silhouette. I can choose to break those up into non-contiguous sub meshes, which look "chunky" when not given enough topology agaisnt their background. Or I can choose to encapsulate them all into as much of a contiguous mesh as possible, which looks fantastic, until you bother to turn on wireframe. Wireframe won't be seen in game, and the mesh is static - so why not use option B, aside from tri-count concerns, and what people will think of it as a portfolio piece? 

poopipe

Assuming ue4 uses similar algorithms to simplygon it'll probably handle a contiguous mesh better than a separated one - you need to have non shit UVs etc. to help it out etc of course.

In this case I'd go with something more like the low poly in the middle - I'd model the silhouette of the recesses and I'd manually lod it because it'd take 5 minutes to do a better job than the auto lodder

I fail to see any difference in approach to texturing or UVs between the two low poly meshes.  They're both cylinders and can be unwrapped as such.

Shrike

I think the middle is the better one, because the last will cause very heavy creases and shading errors in the cylinder when decimated too much

Also, I havent tried U4 lod but the auto LODs ive tried in unity to suck hard with cylinders in general, they work great for things like cars or organic stuff

here is the LOD stack from cinema 4D in this case

Kanni3d

Its completely situational from asset to asset. Some elements on an asset would work best with a contiguous mesh, others would benefit from having their details split off for easier times baking (match by name, isolate ao etc.)

DirtyBlueFL said:

Wireframe won't be seen in game, and the mesh is static - so why not use option B, aside from tri-count concerns, and what people will think of it as a portfolio piece? 

Kinda answered yourself here, if its for portfolio, use whatever looks best, and go for it. Typically portfolio renders do show technical breakdowns, so you'd see the wires anyway.

But technically speaking, aside from tri count, middle would be best in most applications. It appears and is more resourceful, much easier to bake details, contiguous mesh/one, LODing, etc.