A maintenance release fixing latent gradient bugs in less-common polarimetric imaging modes — standard total-intensity (I) and linear-polarization (P) imaging were unaffected.
mcv/vcvpolarization transforms: restored the missing off-diagonal Jacobian cross-term to fix the gradient when imaging circular polarization (Stokes V ≠ 0) together with linear polarization.- Simultaneous Stokes I + polarization imaging: the Stokes-I
log-transform Jacobian is no longer overwritten by the polarization gradient. - Polarimetric TV regularizer: corrected the first-row/column boundary masking in the
ptvgradient. - Crashes fixed: the
mfractional-polarization data term and themsimple/vfluxpolarimetric regularizers when solving for I and polarization simultaneously - NumPy ≥ 1.24 compatibility: diagonalized closures (
cphase_diag,logcamp_diag) build again. - Tests / CI: added a minimal gradient-validation suite (analytic vs finite-difference, NFFT vs direct) running on Python 3.8 and 3.11.
Full changelog: v1.3.1...v1.3.2