femartip/zod-faster-rcnn-metafeatures

ZOD validation meta-features + Faster R-CNN targets

Meta-features extracted from ZOD validation scenes with Faster R-CNN quality targets.

Source

• —Extracted from the validation split of the ZOD frames dataset
• —One row per scene/frame
• —frame_id keeps the original ZOD frame identifier

Contents

• —iou: mean IoU for the scene
• —lrp: LRP for the scene

Use either target column downstream depending on the assessor task.

Columns

• —frame_id
• —country
• —time_of_day
• —lat
• —long
• —road_type
• —road_condition
• —weather
• —solar_angle_elevation
• —month
• —hour
• —forward_acceleration
• —lateral_acceleration
• —forward_velocity
• —lateral_velocity
• —field_view_horizontal
• —camera_distance_from_ground
• —camera_pitch_angle
• —distortion_magnitude
• —camera_offset
• —laplacian
• —quality
• —brightness
• —noisiness
• —sharpness
• —contrast
• —complexity
• —temperature_2m
• —relative_humidity_2m
• —rain
• —snowfall
• —cloud_cover
• —cloud_cover_low
• —cloud_cover_mid
• —wind_speed_10m
• —weather_code
• —num_detections
• —max_conf
• —min_conf
• —mean_conf
• —std_conf
• —iou
• —lrp

Notes

• —The tabular features are the meta-features computed in this repository.
• —The detector-specific datasets include both detector quality targets so that you can choose the target metric later.