Function image_dims_from_shape 

pub fn image_dims_from_shape(
    format: PixelFormat,
    shape: &[usize],
) -> Result<(usize, usize)>

The logical (width, height) an allocation shape describes for format – the inverse of PixelFormat::allocation_shape.

Used wherever a tensor arrives carrying a shape and a format but not its dimensions: the C constructors, the Python bindings and the cross-package protocol import.

The input is the allocation shape, not the addressing shape

PixelFormat::allocation_shape is what Tensor::image allocates from and what descriptor_pinned emits from shape(), so that is what this inverts. PixelFormat::addressing_shape is a different, smaller grid, and for a semi-planar format the two collide: NV12’s addressing shape is [h, w] and its allocation shape is [combined_h, w], both rank 2. Handed an addressing shape this cannot tell – NV12 [480, 640] reads as a combined height of 480 and answers (640, 320), silently. Nothing in the shape distinguishes them, so callers must pass the allocation shape.

A D3D11 import is the one place this cannot bite: the constructors check the derived size against the texture description, so a mismatch fails with Error::InvalidArgument naming both. On Windows, prefer d3d11_texture_geometry and d3d11_shared_handle_geometry, which read the dimensions off the texture itself and leave the shape as a consistency check – except for a semi-planar texture, whose width is its row stride rather than the image width, so those helpers take the width from the shape they are given and refuse to answer without one.

The rank is part of the format’s contract, so a shape of the wrong rank is an error rather than an interpretation of whatever dimensions are present:

Layout	Shape	Result
Packed	[h, w, c]	(shape[1], shape[0])
Planar	[c, h, w]	(shape[2], shape[1])
SemiPlanar	[combined_h, w]	(shape[1], image height)

A semi-planar shape carries the combined plane height, so the logical height is recovered through PixelFormat::image_height_from_combined, which rejects a combined height no image height produces (721 for NV12, say) instead of rounding to a nearby one.

Errors

Error::InvalidShape for a shape of the wrong rank, or a semi-planar combined height with no image height behind it.

Example

use edgefirst_tensor::{image_dims_from_shape, PixelFormat};
assert_eq!(image_dims_from_shape(PixelFormat::Rgba, &[480, 640, 4]).unwrap(), (640, 480));
assert_eq!(image_dims_from_shape(PixelFormat::PlanarRgb, &[3, 480, 640]).unwrap(), (640, 480));
assert_eq!(image_dims_from_shape(PixelFormat::Nv12, &[722, 640]).unwrap(), (640, 481));
assert!(image_dims_from_shape(PixelFormat::Nv12, &[721, 640]).is_err());