Citation

A pimm result normally needs three kinds of citation:

1. the exact pimm software release/commit;

2. every model/method implementation used by the config;

3. the dataset and detector/simulation sources used to create the events.

Do not cite only the repository when the scientific claim depends on a method or dataset paper.

Cite the exact software

This repository does not currently contain CITATION.cff or a canonical software DOI. Until the software-citation TODO below is completed, cite the release URL and commit used by the experiment and retain the resolved config.

uv run python -c 'from importlib.metadata import version; print(version("pimm"))'
git rev-parse HEAD
git status --short

Use this explicitly non-canonical template only when your venue requires BibTeX; replace every bracketed value:

@software{pimm_[release_or_commit],
  author  = {DeepLearnPhysics contributors},
  title   = {Particle Imaging Models (pimm)},
  version = {[release or full commit]},
  year    = {[release year]},
  url     = {https://github.com/DeepLearnPhysics/particle-imaging-models/tree/[tag-or-commit]},
  note    = {Git commit [full SHA]}
}

TODO

Add CITATION.cff, named authors/ORCIDs, a preferred software title, release archive/DOI, and a stable BibTeX entry. Replace the template above once those values are approved.

Find method and dataset citations

Config family	Source of truth to consult	Current documentation status
Panda / Sonata	linked model card, recipe comments, and the implementation lineage in pimm/models/sonata/	canonical paper/BibTeX is not recorded centrally in this repository
Panda Detector	model card plus detector/backbone sources	see TODO below
PoLAr-MAE	PoLAr-MAE source repository and released model card	pimm configs preserve compatibility, but this repository has no central citation manifest
Point Transformer V3 backbone	Point Transformer V3 paper and upstream implementation	cite it when the selected config builds a PTv3 backbone
PILArNet-M	canonical dataset card and its linked publication	license/citation should be copied into a future versioned data manifest

TODO

Publish a reviewed citation mapping for Panda Detector and other research configs that do not yet have canonical method metadata in the repository.

If a source does not provide citation metadata, mark it as pending in your internal record and resolve it with the model/dataset owner before publication. Do not infer a citation from a similarly named project.

Record the configuration behind a result

Alongside the bibliography, preserve:

• pimm version, full commit, and uncommitted diff status;

• resolved_config.json, source snapshot, and exact command;

• checkpoint repository/file plus immutable revision and checksum;

• dataset repository, revision, split, selection, and file manifest;

• preprocessing, coordinate/feature units and order, target class order;

• metric definition, aggregation, evaluation config, and prediction artifact.

The Reproducibility guide gives the full artifact contract.

Attribution and license

pimm is distributed under the MIT License. Vendored or adapted components can carry their own license and citation requirements; check their source directories and upstream projects. Scientific citation is still required when code reuse is permitted by a software license.

Before submitting a paper

• software release and commit are immutable and reported;

• method/backbone/task-head citations match the actual resolved model;

• dataset/simulation citation matches the exact revision and split;

• every pending placeholder on the relevant model/data page is resolved or explicitly disclosed;

• reported metrics can be traced to a saved evaluation config and artifact.