simvx.graphics.assets.cubemap_loader

Load cubemap textures from 6 face images or equirectangular HDR.

Module Contents

Functions

decode_rgbe	Decode a Radiance RGBE / .hdr image to a float32 (H, W, 3) array.
gradient_cubemap_faces	Synthesize 6 cube faces with a vertical gradient from top to bottom.
equirect_to_cubemap	Project an equirectangular (H, W, 3) HDR image onto 6 cube faces.
load_cubemap	Load a cubemap from 6 face images, an equirect HDR, or a solid colour.

Data

__all__
log

API

simvx.graphics.assets.cubemap_loader.__all__

[‘load_cubemap’, ‘decode_rgbe’, ‘equirect_to_cubemap’, ‘gradient_cubemap_faces’]

simvx.graphics.assets.cubemap_loader.log

‘getLogger(…)’

simvx.graphics.assets.cubemap_loader.decode_rgbe(data: bytes) → numpy.ndarray

Decode a Radiance RGBE / .hdr image to a float32 (H, W, 3) array.

Supports both uncompressed (FORMAT=32-bit_rle_rgbe) headers and the standard run-length-encoded scanline form (the only forms produced by every common HDR exporter: Lightroom, Substance, sIBL Archive). No Pillow / imageio / OpenEXR dependency.

Raises ValueError if data is not a recognisable RGBE stream.

simvx.graphics.assets.cubemap_loader.gradient_cubemap_faces(top_rgb: tuple[float, float, float], bottom_rgb: tuple[float, float, float], size: int = 64) → list[numpy.ndarray]

Synthesize 6 cube faces with a vertical gradient from top to bottom.

Returns 6 (size, size, 4) float32 RGBA arrays in Vulkan order [+X, -X, +Y, -Y, +Z, -Z], values in [0, 1]. The gradient is sampled by each texel’s world-space direction (its y component) so the horizon ring is continuous across all faces: texel-row gradients kink at the seams.

Math mirrors the web _gradient_cubemap_faces (web/renderer/web.py) exactly so both backends produce an identical gradient sky / IBL ambient. The two are kept as parallel impls because the web renderer runs in Pyodide and cannot import this module (cubemap_loader imports vulkan); a future single-source would move the pure-NumPy core into simvx.core.

simvx.graphics.assets.cubemap_loader.equirect_to_cubemap(equirect: numpy.ndarray, face_size: int = 256) → list[numpy.ndarray]

Project an equirectangular (H, W, 3) HDR image onto 6 cube faces.

Returns a list of 6 (face_size, face_size, 4) float32 RGBA arrays in Vulkan cubemap order [+X, -X, +Y, -Y, +Z, -Z]. Sampling is bilinear in the source image; corner pixels stay numerically stable because the direction vectors are normalised before projection. Pure-numpy: vectorised over every pixel of every face for sub-second turnaround at 256² faces.

simvx.graphics.assets.cubemap_loader.load_cubemap(device: Any, physical_device: Any, queue: Any, cmd_pool: Any, face_paths: list[str] | None = None, hdr_path: str | None = None, colour: tuple[float, float, float] | None = None, face_size: int = 256, faces: list[numpy.ndarray] | None = None) → tuple[Any, Any, Any, Any]

Load a cubemap from 6 face images, an equirect HDR, or a solid colour.

Args: face_paths: List of 6 image paths [+X, -X, +Y, -Y, +Z, -Z]. hdr_path: Single equirectangular .hdr file (Radiance RGBE). Projected onto a 6-face cubemap on the CPU; face_size controls the per-face resolution (default 256). colour: Solid colour (r, g, b) in 0-1 range (fallback if no paths). face_size: Resolution of each cube face when projecting from an HDR equirect. Ignored for explicit face_paths.

Returns: (image_view, sampler, image, memory) tuple for the cubemap.