Has anyone seen a texture result similar to the following image, where some of the faces are drawn with what appears to be non-contiguous (and wrong) uv assignments?
If this is known problem, are there any workarounds?
In the above case, my steps were: import images, align, reconstruct at high resolution, unwrap with the legacy setting, and texture. I didn’t export/re-import the model, the result is purely staying within the RC pipeline.
This looks related to a recent post showing similar results for a model that was imported to RC for texture projection, and the advice in that case (adjust textel size in unwrapping) may be related here in the all-RC workflow but I haven’t been able to verify that modifying those values can help here.
I agree with Gotz here, I’d spike the large triangle removal threshold and re-texture. I large triangle number is somewhat relative, but I still spike it to 100 or even 500 depending on the scene (latter is if I’m trying to get some detail out of a skydome or plane projection).
Thanks for the fast and helpful notes – below is a twist I didn’t convey above but feel you might have some good thoughts on.
First, increasing the ‘large triangle removal threshold’ from the default value of 10 does help by including the large triangles shown above in the map, which are otherwise ignored.
But what is interesting is that as this value increases in poorly sampled regions I see exactly the opposite effect described above.
Below, compare ‘large triangle removal threshold’ results with values of 50 and 500. The higher value shows the large artifacts! Again, this is only true for weakly sampled areas where there are fewer images to poll to draw the texture. Perhaps this is because larger triangles can be considered, and as a consequence poorly sampled areas now contribute more artifacts to the texture. Perhaps something else is going on here entirely.
As shown at top, the region with the large poster-color polys is covered well in the UV map.
If I find the time I’d like to reproject from the corrected images in other software to see what these regions look like there – my sense is that this is an artifact in the way RC traverses the uv map during texturing, and that a brute force approach to reprojection would not show these.
