Configuration
Event-LAB reads its main settings from config.yaml in the project root.
Most users only need to change the frame generation and evaluation settings
shown below.
Evaluation settings
ground_truth_toleranceDistance in metres used when building the ground truth tolerance around a matched reference place. A value of
3means a prediction within 3 m of the ground truth match is counted as correct.filter_places_secMinimum time in seconds between frames used as places. Set this to
0.0to keep every generated frame. Increase it when you want fewer, more separated places along the route.
Frame generation settings
timewindowsList of time windows, in milliseconds, used when generating frames by time. Event-LAB runs the baseline once for each value.
num_eventsList of maximum event counts per frame. This is only used when the selected frame generation mode is event-count based.
frame_generatorControls the type of frame data prepared for the baseline.
framesBuild accumulated event frames directly from the event stream.
reconstructionBuild image-like reconstructed frames using the selected reconstruction model.
frame_accumulatorControls how events are accumulated into direct event frames.
eventcountAccumulate by event count.
polarityKeep polarity-aware event information.
reconstruction_modelReconstruction model used when
frame_generatorisreconstruction. Current options aree2vidandfirenet.
Example
ground_truth_tolerance: 3
filter_places_sec: 0.0
timewindows: [250, 500, 750, 1000]
num_events: [100000]
frame_generator: "reconstruction"
frame_accumulator: "polarity"
reconstruction_model: "e2vid"
With this configuration Event-LAB will generate reconstructed frames at four time windows and run the selected baseline once for each window.