A question about low poly modeling.

Geekson

Should I weld separate meshes in low poly model? The pros and cons of these approaches? In which cases I should weld, and in which cases I shouldn't do that? 
And some images:

rohMizuno

If is a hard surface piece like those it is okay to just have one sitting over the order and save a couple of triangles by not welding it together, in most cases.
Sometimes you might want to make it all one piece, for example if the angle in between both piece is higher than 90 degrees. That way you can keep those UVs together in order to avoid seems and to make you life easing when texturing (if the angle is 80 or lower you will probably cut the UVs anyways, so it does not meter).
But keep in mind that the geo you have "hidden" under the other piece will be dead space on your UVs, so its not a bad idea to some times run a cut and delete those, it all depends if you value tri count or texture space on the model.

Scruples

Some Z fighting/Z-buffer errors will occur too, not to this degree but something to be aware of, and it only gets worse as distance to the object increases.

Mark Dygert

It really depends on the object, it's importance and what you're trying to do. 

It's not always about saving a few polygons, sometimes its about UV layout optimization, shading or a better workflow.

For example, the the UV layout in the top pic would be a giant quad with a big deadzone in the middle, that uv space is wasted. If you try to save even more poly by clipping the cyclinder through a box, you waste even more UV space and risk shading and zfighting like Scruples mentioned. You can use those dead areas but it will be fighting with whatever is underneath. 



If the pieces are conjoined your UV layout has a hole that you can put other pieces into and they won't show up or shade on anything else. 
If you have vertex shading you need verts on the plane at the base of the cylinder so the shading on the plane works with the shading on the cylinder. If you don't, it leaves the 4 verts at the corners of the quad to properly shade around the cylinder. But if you're doing another type of lighting or it's full bright/hand painted, it might not matter that much, so it depends. 

Left, joined. Right Clipping.
The vert in the center of the top cylinder will usually shade dark in most instances because it's hidden by the lower cylinder. If you remove it, then there isn't anyway to shade the bottom of the upper cylinder, except for the outer verts on the edge. If you don't have this type of shading it might be ok to clip them together. That happens more often on super low poly hand painted mobile games.

Low poly has become a lot higher poly in recent years and things like bevels and slopes have become almost vital, especially when baking textures. So for example you want to transition some material detail from the plane to the cyclinder, like a weld bead where they join together, half of the weld on the plane, half on the cylinder. Or some damage or a stripe that goes across both, the transition gets easier if they are contiguous. 



For any deforming mesh, you almost always want them to be contiguous, otherwise the point that they meet will deform differently, tearing away and whatever is being covered up will come poking out. 

An AO bake shows the dark shadows poking out from under the spikes when the meshes deform. If they are welded, they share similar weights and both deform the same way.

Bek

still relevant: http://polycount.com/discussion/117017/question-about-continuous-meshes-for-low-poly-bakes

Geekson

rohMizuno said:

But keep in mind that the geo you have "hidden" under the other piece will be dead space on your UVs, so its not a bad idea to some times run a cut and delete those, it all depends if you value tri count or texture space on the model.

Thanks, I didn't think about UV wasting, that's a good point.

Scruples said:

Some Z fighting/Z-buffer errors will occur too, not to this degree but something to be aware of, and it only gets worse as distance to the object increases.

Now I know what it is Thanks!

Mark Dygert said:

It really depends on the object, it's importance and what you're trying to do. 

It's not always about saving a few polygons, sometimes its about UV layout optimization, shading or a better workflow.

For example, the the UV layout in the top pic would be a giant quad with a big deadzone in the middle, that uv space is wasted. If you try to save even more poly by clipping the cyclinder through a box, you waste even more UV space and risk shading and zfighting like Scruples mentioned. You can use those dead areas but it will be fighting with whatever is underneath. 

Many thanks for the detailed answer! I think that I don't have any questions now

Bek said:

still relevant: http://polycount.com/discussion/117017/question-about-continuous-meshes-for-low-poly-bakes

Great, I haven't seen that topic, really helpful, thanks a lot!

Geekson

And another question: how is better merge object with cylindric shapes? I had some normal map troubles with smoothing in my previous study projects, and now I want to understand what causing them, and how to avoid it.  I believe in that particular case I can just don't merge cylinder to other objects, but what if it must be done? 
And this red arrow - is it acceptable to merge object like that?

musashidan

The area where I've drawn the green tri is how that quad will triangulate so that is fine. The green tri below that will triangulate like that(which is also fine) Tris to avoid are shaped like the one in the dotted red line: they are the same shape as the large green tri but with a much lower peak. Also, tris that are long and extremely thin can cause issues with both baking and rendering in-engine. These tris can often result if you try to join curved topology with a lot of segments/edges to a sparse surrounding topology, which yields high valence poles(verts with a high amount of converging edges)

The geo I've marked with the blue box will be difficult to bake, in the sense that the projection rays are so closely packed that they are sure to intersect(even if using a cage or averaged bake) For a case like that you might be better filling in the geo either flat or with chamfering/shallow extrudes, and allowing the normal map to do the rest of the work.

Your edge termination approach is correct and should be applied wherever it suits on the mesh to reduce more dense topo  as it connects to less dense topo. Care should be taken when doing this that the all-important silhouette is maintained.



Personally, for hard-surface assets I will usually separate the components of the model where it makes sense.(where it might be separate in the real-world due to the manufacturing method)

Trying to always connect topo together into a contiguous mesh isn't always the best approach. This will, of course, depend and vary from asset to asset and indeed on parts of the same asset. These are all calls you will eventually be able to judge for yourself the more experience you have bringing assets through the pipeline from modeling to finished in-engine.

So all of the great points mentioned above like z-fighting/vert count/uv space/etc, may or may not be an issue from asset to asset but knowing about them is very important as you will have the knowledge of why certain problems arise as they occur and, more importantly, what to look out for/do about it, if they do.