LEMURS: Large-scale multi-detector ElectroMagnetic Universal Representation of Showers

The LEMURS dataset is described in details in arXiv:2509.05108.

Photon (gamma) showers are presented, developping in 5 different electromagnetic detectors (Par04SiW, Par04SciPb, ODD, FCCeeCLD, FCCeeALLEGRO) using custom scoring mechanism (Universal Representation of Showers) . Nearly 1M showers per detector are simulated in a Universal Representation, which was implemented first in Par04 extended example of Geant4, and used also by dataset 2 of CaloChallenge.  A file structure is slightly changed with respect to CaloChallenge files. Wide range of energies and angles are provided. Energy range differs for experiments on electron-positron colliders (FCCee*, up to 100 GeV) and for hadron colliders (Par04* and ODD, up to 1 TeV).

For each of the five detectors 10 HDF5 (.h5) files are provided for the main/training dataset (up to 100k per file), and a zip archive with 8 files per detector for the smaller testing (discrete) datasets.  Training files are not grouped in an archive as they are large and we wanted users to have the flexibility to choose the number of events they download. Each HDF5 files contains following structure:

Validation plots (at least one per file) are included in the publication.

Additionally to the dataset training and testing files, LEMURS logo is provided (logo.zip) and can be used when presenting results or referencing the dataset to facilitate recognition.