Data Filtering
Data Filtering helps users refine research materials. It enables efficient selection of the most relevant samples from large datasets based on attributes, labels, or research objectives. Whether finding observations under specific conditions or excluding outliers to improve training quality, it provides flexible filtering options for high-quality, targeted analysis.
Perception
For Perception, dataset filtering is mainly based on variables such as weather, sensors, and traffic scenarios.
Filter data from different scenes in the raw dataset.
"all": all data by default
"rain": obtain the subset of data for rainy conditions
"snow": obtain the subset of data for snowy conditions
"fog": obtain the subset of data for foggy conditions
"brightness": obtain the subset of data for bright-light conditions
"darkness": obtain the subset of data for nighttime conditions
Filter data from different scenes in the raw dataset.
–"intersections": obtain the subset of intersection data
–"unsignalized intersections": obtain the subset of signalized-intersection data
–"signalized intersections": obtain the subset of unsignalized-intersection data
–"highway": obtain the subset of highway-segment data
–"ramp": obtain the subset of ramp merge/diverge area data
–"roundabout": obtain the subset of roundabout data
Filter data from different sensors in the original dataset.
–"Camera": obtain the camera-image data subset
–"LiDAR": obtain the LiDAR point-cloud data subset
–"RADAR": obtain the radar point-cloud data subset
–"IMU": obtain the inertial measurement unit data subset
–"GPS": obtain the Global Positioning System data subset
Prediction
For Prediction, dataset filtering is mainly based on traffic scenario, interaction conditions, and traffic-flow density.
Filter data from different scenes in the original dataset.
–"intersections": obtain the subset of intersection data
–"unsignalized intersections": obtain the subset of signalized-intersection data
–"signalized intersections": obtain the subset of unsignalized-intersection data
–"highway": obtain the subset of highway-segment data
–"ramp": obtain the subset of ramp merge/diverge area data
–"roundabout": obtain the subset of roundabout data
Filter data for different interaction objects in the raw dataset.
–"motor-motor": obtain scene data for vehicle-to-vehicle interactions
–"motor-non_motor": obtain scene data for motor vehicle–non-motor vehicle interactions
–"motor-human": obtain scene data for vehicle–pedestrian interactions
Filter data with different traffic-flow densities in the raw dataset.
–"high": obtain high-density traffic-scene data
–"medium": obtain medium-density traffic-scene data
–"low": obtain low-density traffic-scene data