Event Display

Live: huggingface.co/spaces/CERN/colliderml-event-displaySource: spaces/event-display/

Interactive 3D viewer for single events from the ColliderML datasets. Pick a process and an event ID to see the tracker hits, reconstructed tracks, and truth particles inside the OpenDataDetector geometry — rendered with Plotly so you can rotate, zoom, and toggle layers.

What you can see

• Blue points — tracker hits, coloured by detector layer (or volume_id / particle_id if layer is missing).
• Red lines — reconstructed tracks, approximated as helical polylines from the perigee parameters (d0, z0, phi, theta, qop). This is a visualisation aid, not a physics-accurate reconstruction.
• Yellow dashes — truth-particle momentum vectors drawn from the primary vertex, restricted to primary particles and capped at the 20 highest-momentum entries so the plot stays readable.

Datasets

The dropdown ships with a curated subset of the catalogue so the UI stays responsive on the Space's modest resources:

ttbar_pu0
ttbar_pu200
higgs_portal_pu10
zmumu_pu0
diphoton_pu0
single_muon_pu0

Each selection loads at most 50 events into memory. If you want a different process, open an issue — or run the Space locally and edit DATASETS in app.py.

Data loading

The Space prefers its own on-disk cache (_cached_events/<dataset>/*.parquet) if present; cache_events.py populates it by calling colliderml.load() once for each dataset and writing the results to parquet. On a fresh Space with an empty cache the app falls back to calling colliderml.load() live from the dataset's HuggingFace repo — slower on first hit, then cached in memory via functools.lru_cache.

Running it locally

cd spaces/event-display
pip install -r requirements.txt
python app.py
# Open http://localhost:7860

Customisation

• Add a dataset → append to DATASETS in app.py and re-run cache_events.py.
• Add a visualisation layer (e.g. calo cells) → follow the pattern for tracker_hits in render_event() and add a new colliderml.load table to the list in _load_dataset().
• Change the colour scheme → the Plotly Scatter3d traces carry all the knobs; search for Viridis and crimson in app.py.