Datasets

Event-LAB datasets are selected by name from the command line:

pixi run eventlab <baseline> <dataset> <reference> <query>

For example:

pixi run eventlab sparse_event brisbane_event sunset2 sunrise

This loads datasets/brisbane_event.yaml, downloads the sunset2 and sunrise sequences if needed, formats them into Event-LAB’s standard HDF5 format, generates frames using config.yaml, and prepares the ground truth matrix for evaluation.

Available datasets

The dataset name is the YAML filename in datasets/ without the extension. The sequence names are the keys under sequences in that YAML file.

Dataset	Command name	Example sequences
Brisbane-Event-VPR	`brisbane_event`	`sunset1`, `sunset2`, `sunrise`
NSAVP	`nsavp`	`R0_FA0`, `R0_FS0`
Fast-and-Slow	`fast_slow`	`r_med1`, `q_med1`
QCR-Event-VPR	`qcr_event`	`normal1`, `fast1`, `slow1`

Download data only

To download and prepare the datasets listed in config.yaml without running a baseline:

pixi run getdata config.yaml

The datasets section in config.yaml controls which dataset and sequence pairs are prepared.

Dataset YAML files

Each dataset YAML file describes:

dataset: Metadata such as the dataset name, camera, resolution, and source URL.
sequences: Download links for each sequence’s event data, optional ground truth, and optional hot-pixel file.
format: Input data format and timestamp scale used during conversion.
other: Dataset-specific details such as sequence offsets or manual ground truth timing points.

The Brisbane dataset file is the best starting point when checking how a full dataset configuration should look.