Sabbysab/a2d2-front-center-subset

A2D2 Front-Center Camera Subset

A subset of 1,978 front-center camera frames with their semantic segmentation label masks from the semantic segmentation part of A2D2. Frames are sampled evenly from all 23 annotated scenes (recorded in southern Germany).

It is used by Zero-Shot Driving Scenario Search to display search results.

Attribution and license

A2D2 – Audi Autonomous Driving Dataset, © Audi AG. Licensed under Creative Commons Attribution-NoDerivatives 4.0 International (CC BY-ND 4.0).

• Original source: https://a2d2-dataset.github.io (public bucket audi-autonomous-driving-dataset, camera_lidar_semantic/)
• No modifications: every image and label mask is the unmodified original file. Only the selection of frames and the folder layout differ.
• Citation: Geyer et al., A2D2: Audi Autonomous Driving Dataset, arXiv:2004.06320, 2020.

Structure

Path	Content
camera/<scene>/<scene-id>_camera_frontcenter_<frame>.png	Original RGB frame (1920×1208)
label/<scene>/<scene-id>_label_frontcenter_<frame>.png	Original semantic label mask
class_list.json	Original A2D2 color → class mapping
manifest.csv	scene, image_path, labels: classes covering ≥ 1,500 pixels in each mask (numbered variants such as "Car 1"–"Car 4" are grouped as "Car"). This is project metadata derived from the masks. It is not part of A2D2