Calibration¶
Vendor-neutral estimation problems that turn raw MRIData into the inputs an
operator (Layer 4) needs — currently just coil sensitivity maps, used by
reconstruct(method="cg") when sensitivity isn't supplied explicitly. See
Architecture → Calibration & preprocessing.
estimate_sensitivity(method="rss") is a crude, dependency-free baseline:
per-coil images from a single-pass gridding/iFFT (the same coil_images()
building block reconstruct(method="adjoint") uses), normalized by their
root-sum-of-squares combination. It is not a substitute for ESPIRiT or an
adaptive/Walsh combine — those are planned (Milestone 4, wrapping sigpy.mri
/ BART rather than reimplementing them) and will be selectable via the same
method= argument.
Return coil sensitivity maps, shape (n_coils, *image_shape).
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
data
|
MRIData
|
Multi-coil :class: |
required |
method
|
str
|
|
'rss'
|
eps
|
float
|
Regularization floor for the RSS normalization, relative to its max. |
1e-06
|