Handle alpha in textures? about xatlas HOT 14 CLOSED

Seems like I can modify https://github.com/jpcy/xatlas/blob/master/extra/example_uvmesh.cpp#L83 to handle the 4 8-bit colors and 4 float colors.

from xatlas.

Apart from the repack example (which is incomplete), the other example output images are just for visualization. The rasterization used to generate them isn't precise, so I don't recommend using them for anything.

from xatlas.

My goal is to take multiple meshes make them use 1 texture and uv space. Then, merge the meshes.

Is this not suitable?

from xatlas.

The repack example here does that, but it doesn't handle bilinear filtering yet.

from xatlas.

Why would it need to handle bilinear filtering?

from xatlas.

I have access to:

INTERPOLATE_NEAREST
INTERPOLATE_BILINEAR
INTERPOLATE_CUBIC
INTERPOLATE_TRILINEAR
INTERPOLATE_LANCZOS

from xatlas.

Because bilinear filtering samples neighboring texels. The repack example copies texture data by conservatively rasterizing charts in the new UV space. A source texel on the edge of a texture may end up in the middle of the atlas and bilinear filtering would lerp with neighboring black texels.

from xatlas.

If we made the original texture twice as big using a filter like LANCZOS and would it have the same effect?

Or the other way around shrinking by half...

from xatlas.

Going back to the theory. Sorry for my misunderstandings.

from xatlas.

Why can't we pad the charts with suitable pixels? https://github.com/jpcy/xatlas/blob/master/xatlas.h#L204

from xatlas.

I have access to a function that takes a square size of image data and resizes it.

template <int CC, class T>
static void _scale_lanczos(const uint8_t *__restrict p_src, uint8_t *__restrict p_dst, uint32_t p_src_width, uint32_t p_src_height, uint32_t p_dst_width, uint32_t p_dst_height)

You create a bounding rectangle here https://github.com/jpcy/xatlas/blob/master/extra/example_repack.cpp#L359.

Not sure where to go.

from xatlas.

@fire you can check AliceVision on github to see how it handles bilinear filtering. It uses geogram which uses xatlas. Here, this is how you do it. I'm going to implement it for example_repack in a week or so.

from xatlas.

There shouldn't be any artifacts with bilinear filtering now.

Comparing input to output, they don't quite match and there's some distortion in the texture data. I think something is off by half a texel.

from xatlas.

Trying the revised code today.

from xatlas.