multigeom.cli

Command line: multigeom <command> [options] (or python -m multigeom.cli). Each command hands its options to the module that does the work, so multigeom <command> –help lists them.

jacobian         the Fisher metric J(x) of an encoder, for some test images
compute-g        the multi-scale metric G(x) (and J(x)), resumable
train-diffusion  the conditional diffusion model on precomputed responses
align            alignment and Bures-Wasserstein distance between configurations
train-encoder    fit an encoder readout to neural data (needs the [paper] extra)
responses        precompute the encoder responses on TinyImageNet (core)
metric-field     the local metric field of Fig. 5

Functions

jacobian([argv])	J(x) = J_f^T J_f / sigma^2 of an encoder: its top eigenpairs for one image, or the first N test images.
main([argv])	Run multigeom <command>; returns the exit code.