LEMURS-ds3 dataset: Large-scale ElectroMagnetic Universal Representation of Showers with higher granularity

This dataset (LEMURS-ds3) is a follow-up of LEMURS dataset, but using a single detector (Par04 SiW), but with a higher granularity. The structure of the dataset is identical to LEMURS, with the dimension of Universal Representation of Showers changed from (R,Phi,Z) = (9,16,45) to (R,Phi,Z)=(18, 50, 45). This means that radially the voxels are twice smaller, and in azimuthal angle the full 2*pi is divided by 50 instead of 16 voxels.

This granularity is an equivalent of CaloChallenge dataset3 (hence the name). The same detector was used in both cases, here photon showers are presented, and in a much larger scale (1 million showers) and much wider region of the detector (theta, phi).

The structure of the HDF5 file is as follows:

10 files of the training datasets contain 100k showers each (initiated by incident photons).

The zip archive contains additionally 2 discrete testing datasets (`testing/`) with 1k showers that may be used for final validation, with validation plots as described in detail for LEMURS dataset in its description on arXiV within `plots/` directory (both for testing and training datasets).