README.md

metadata

license: unknown
task_categories:
  - object-detection
tags:
  - physics
  - liquid-argon
  - particle-imaging

PILArNet-M-mini

A tiny 120-event subset of DeepLearnPhysics/PILArNet-M, stored in the exact same HDF5 format (variable-length point/cluster/cluster_extra datasets).

• train/generic_v2_80_v2.h5 - 80 events
• val/generic_v2_20_v2.h5 - 20 events
• test/generic_v2_20_v2.h5 - 20 events

Intended purely for CI/CD and exploration - a lightweight drop-in for pipelines that expect PILArNet-M without downloading the full multi-TB dataset.

Revisions

main tracks the v3 layout of the parent dataset: cluster is 6 columns and cluster_extra is 6 columns ([_, momentum, vtx_x, vtx_y, vtx_z, is_primary]), with vertices in the same voxel frame as the point coordinates.

The previous v2-layout mini (5-column cluster_extra) is preserved on the v2 branch.

Events are the first 80/20/20 events of each split's parent shard with at least 1024 points.