CLI reference¶
The preferred public command is:
The subcommands below were checked against the current toolkit interface on 2026-09-05. Re-check them before public release if the toolkit source or package name changes.
Current subcommands¶
hemispec quickstart run the public-safe synthetic smoke test
hemispec models list or pre-download released trained DGN model bundles
hemispec infer run trained DGN inference on preprocessed GM maps
hemispec compute compute ANS/RNS maps from actual and reconstructed GM maps
hemispec run run DGN inference followed by ANS/RNS computation
hemispec workflow run bilateral DGN and generate ANS/RNS outputs
hemispec trt test-retest reliability validation
hemispec specificity structural specificity validation
hemispec hemi-classify ROI-level hemisphere-classifier validation
Command naming¶
Use hemispec for the command-line interface and hemispec-gui for the graphical interface.
Model asset prefetch¶
Current main can pre-download the DGN checkpoints tracked by Git LFS:
If this is skipped, workflow, infer, run, and the GUI download the DGN checkpoints from the repository's main Git LFS media on first model use.
The CLI still supports --with-classifier, but it is not a working installation recommendation at present because required feature_names.csv media URLs return HTTP 404. Obtain classifier assets from a local Git LFS checkout and use --classifier-model-dir or HEMISPEC_CLASSIFIER_MODEL_DIR; see the known limitation.
ROI outputs¶
ROI export is currently exposed through options on compute, run, and workflow. For workflow, ROI export is optional and can be skipped with --no-roi-table; classifier validation is opt-in with --run-classifier and requires ROI features.
There is not yet a standalone roi subcommand.
The released classifier accepts only features from its compatible Glasser 1.5 mm atlas: homologous labels 1..180 on the left and 1001..1180 on the right. A custom atlas may be used for ROI export, but not with the released classifier.
Reporting¶
There is not yet a standalone report subcommand. Reporting should be treated as a planned feature until implemented.
ANS/RNS and the cross-hemispheric DGN framework originate from Wang et al. (2024); see Citation.