Detector event views

The reusable asset gallery provides hits, tracks, clusters, particles and overlays for one ttbar event per detector family: CLD, CLIC, IDEA, MAIA and ColliderML. Each family has white PNG, transparent PNG and SVG exports, plus an extracted scene archive for regeneration.

All CLD panels show event 0 from the same 365 GeV ttbar simulation. The particle panel shows Pandora reference candidates. Blue and orange routes identify the inputs and output roles of the two MLPF model families.

Particle sources¶

The gallery records sampling, scale, source collections and trajectory conventions. Use the dataset catalog for model-compatible datasets and the raw-data comparison guide for source provenance.

Hosting and regeneration¶

Visual assets, extracted scene archives and the provenance manifest are hosted in jpata/particleflow on Hugging Face. The documentation embeds those files directly. Generated assets stay under the ignored local_test_data/ directory.

To regenerate from locally available detector data:

PYTHONPATH=. MPLCONFIGDIR=/tmp/pf-mpl .venv/bin/python scripts/create_reconstruction_event_assets.py

To reproduce the published events without the original detector files, download the extracted scenes and render them locally:

hf download jpata/particleflow \
  --include "images/event_displays/*/event_0_scene.json.gz" \
  --include "images/event_displays/manifest.json" \
  --local-dir local_test_data/event_assets

PYTHONPATH=. MPLCONFIGDIR=/tmp/pf-mpl .venv/bin/python scripts/create_reconstruction_event_assets.py \
  --from-scenes --output-dir local_test_data/event_assets/images/event_displays

Use --revision <commit> with hf download to pin an asset version. The manifest records scene SHA-256 hashes, source collections, event identifiers and display settings. Use --source cld=/path/to/file.root to override a raw source, --event to select an event and --detectors to choose families. ColliderML also accepts --colliderml-parquet for saved clusters.