search

KhronosGroup / KTX-Software

KTX (Khronos Texture) Library and Tools
Other
875 stars 230 forks source link

Support equirectangular cubemap projections #814

Open wasimabbas-arm opened 10 months ago

wasimabbas-arm commented 10 months ago

Most of the environemnt cubemaps one can find online are in equirectangular format. Its impractical to expect people to convert those to 6-cubemap images before they can be turned into a ktx file.

IIRC there is a way to extract data for each side from the equirectangular projection onto each cubemap face without having to render it like a 3D engine would do. It will be nice if this is supported.

MarkCallow commented 10 months ago

Its impractical to expect people to convert those to 6-cubemap images before they can be turned into a ktx file

Is it? How do people use these images on GPUs? Do they not make cubemaps?

wasimabbas-arm commented 10 months ago

AFAIK there are a few ways how people would use it.

First options is to create a cube map offline using tools like https://github.com/dariomanesku/cmftStudio. That can read Input and output formats: *.dds, *.ktx, *.hdr, *.tga. and write out Input and output types: cubemap, cube cross, latlong, face list, horizontal and vertical strip. You can then load those as normal. In Vulkan you should have to load each face, each mip-map etc into the GPU.

The other options is that you load the source file in latlong and do a render to all 6 faces of a cube using the texture and generate the cubemap that way. Provides more control and easy to change cubemaps but requires more runtime processing time.

The other options is to just use the latlong texture as it is and instead of doing cubemap filtering do 2D filtering using the formula that, I can't remember and mentioned above. This option gives you even more control but doesn't use the hardware capabilities available.

Pros and Cons: Option 1: Pros: Not ktx-software problem. Compressions is easy, each side is compressed separately. No code required to extract faces. Cons: Pre-procssing required. Multiple tools required to end up with a ktx cubemap. Will be harder in automated workflows.

Option 2: Pros: Can support any projection of cubemap. Cons: More engine work required. Requires runtime cost atleast once. If the latlong image was compressed, your cube map will endup with some uncompressed format or if framebuffer compression available it might be possible to use that.

Option3: Pros: No offline or runtime tools required to do anything. Compression just works. Cons: Not using the hardware capabilities available. (Can't say if it will be faster or slower)

I would really like the first option but without having to use another tool to create those 6 faces first.