Skip to main content

tiled_field_inversion

qsm_core::inversion::tiled

Function tiled_field_inversion 

Source 
pub fn tiled_field_inversion(
    field: &[f64],
    mask: &[u8],
    grid: &Grid,
    model: &OnnxModel,
    divisor: usize,
    cfg: &TileConfig,
    pre: impl Fn(f64) -> f32 + Sync,
    post: impl Fn(f32) -> f64 + Sync,
    progress: impl FnMut(usize, usize),
) -> Result<Vec<f64>, OnnxError>
Expand description

Run a fully-convolutional field→χ ONNX net patch-by-patch with a context halo, bounding peak memory to a single patch.

Values are fed/read as f32 in the x-outer NCDHW layout the nets use; pre maps each input field value, post maps each raw net output value. Each patch is zero-padded to a multiple of divisor (the net’s size_divisor) and to include the halo. The result is masked, matching the whole-volume wrappers.