Contributing Datasets

Add a dataset when Event-LAB can download the raw event data and describe each sequence in a YAML file. The Brisbane configuration is the clearest complete example.

Dataset YAML

Create a file in datasets/ named after the command users should type. For example, datasets/my_dataset.yaml is used as:

pixi run eventlab sparse_event my_dataset reference_sequence query_sequence

The YAML should include:

dataset

Name, version, description, source URL, camera, and resolution.

sequences

One entry per traverse. Each sequence needs a data URL, plus ground truth and hot-pixel entries marked as available or unavailable.

format

Input event data format, timestamp units, ground truth format, and hot-pixel format.

other

Optional dataset-specific information such as per-sequence offsets.

Template

Use datasets/dataset_template.yaml as the smallest starting point. Copy it to a new dataset YAML and replace the names, URLs, formats, and sequence keys.

Ground truth

If ground truth files are available, set ground_truth.available to true and provide the download URL. Event-LAB will download the files and build the reference/query ground truth matrix during a run.

If a dataset does not provide ground truth files, leave the URL empty and set available to false. Datasets that need manual timing information can use the other section, as in datasets/qcr_event.yaml.

Adding the dataset to defaults

After adding the dataset YAML, add an entry to the datasets list in config.yaml only if you want pixi run getdata config.yaml to prepare it by default.