Does anyone have an example of mirrored UVs working (with no seams) with a normal map in Marmoset?

--E--

I can’t seem to get mirrored UVs to work 100% with tangent normal maps.  I’ve tried offsetting the mirrored part outside of the 0-1 UV space.  I’ve tried different tangent spaces in Marmoset and hitting the fix mirrored UVs button in Marmoset and it doesn’t seem to make a difference.  There’s always a slight UV seam no matter what when I crank up the gloss/metalness sliders. 

EarthQuake

Where are you baking the normal map?

Generally speaking, you may see seams at any UV seam if you zoom in far enough or if the texture resolution is relatively low. This comes down to how pixels are mapped to to UV space.

--E--

Hey Joe! Thanks for the reply!

I've tried baking in Marmoset and Painter. I'm baking a 2048 texture onto just a half a sphere.  If it's just not possible to fully get tangent space normal map UV seams to go away on mirrored meshes then that's just how it is but I wanted to hear it from someone who knows better than I, like you! . Here is a link to my files high and low > https://www.dropbox.com/s/q9gfo2pdujqb4zz/mirrored_uvs.zip?dl=1

--E--

Here's a screen shot from marmoset where this map was baked. As I said the seam is pretty subtle, but it's there.

Here's a screen shot from painter where this version of the map was baked, the seam is way more apparent.

EarthQuake

Thanks for the images. Does the issue improve if the texture resolution is increased? If so we'll know it's a resolution problem more so than a mirroring or seam specific problem.

pior

@--E--
 Your low is made of quads, which means that you cannot guarantee that a given baker or engine will interpret the triangulation of it this or that way. It can be okay in some cases (dense organic meshes for instance), but when absolute precision is needed (like here, obviously) you simply cannot afford triangulation not being explicitly controlled.



Now I am not saying that this is absolutely what is causing your issue - but this is a parameter you cannot ignore. This is precisely why asking for examples is not going to get you very far. The only thing that matters is your practical case, as there are more factors involved than it may seem.

And FWIW, to bring the point across : getting the same faint seam myself even with a 100% controlled triangulated model ... which is exactly why splitting UVs things down the middle of important pieces is never a good idea to begin with Not only because of technical reasons, but also because it means that any weathering pass will likely end up being quite noticeably symmetrical ...



One interesting aspect of your example is that while it is not representative of a "real world" scenario (mostly because your low is really quite different from your high) it does reveal a potential issue that is not noticeable in many cases. For instance : 



Fun stuff.

--E--

Thanks for the reply guys!  I really appreciate you guys taking the time to try to help me sort this out!

@Joe - Yeah I've tried 4096 and 8192 maps with 8x, 16x 64x sampling, dithering on and off, soften on and off, smooth cage on and off, fix mirrored tangents on and off, all the tangent spaces padding at all different levels. triangulated mesh. There's always a seam no matter what.

@Pior - Yes, good catch that example wasn't triangulated as it should be.  I have tried it triangulated as with many other simple tests and I still always get a seam. But you are right it should be triangulated before baking.

Edit - I was able to get another quick test example working much better. The seam is extremely hard to find.  but I have zero idea why it's working so much better. It has just as much difference between the high and the low as my last test. But virtually no seam when baking in both marmoset and painter. But zero idea why, and not able to reproduce it on other tests... =/

https://www.dropbox.com/s/g1f0cq1l343hy7u/mirroed_uvs_working_1.8.2020.zip?dl=1

EarthQuake

@--E-- I tried playing around with these scenes and various settings and manually triangulating the mesh and couldn't get a better result. Mapping it uniquely (no mirroring) fixed it. I suspect it may have something to do with what @pior suggested, that if the low poly matched the high a bit better the seam would be less noticeable.

--E--

Thanks for taking the time to try Joe, it means a lot just knowing that it's not some simple setting I'm missing.

Cheers