Symmetric Ambiguity Spacecraft Dataset

Symmetric Ambiguity Spacecraft Dataset (SASD) Overview Symmetric Ambiguity Spacecraft Dataset (SASD) is a synthetic dataset designed for monocular pose estimation of non-cooperative spacecraft. It is built based on the publicly available 3D model of the Symmetric Ambiguity Spacecraft Dataset (SASD) from NASA. Compared with existing datasets such as SPEED, SASD provides richer and more comprehensive supervision, including: 2D semantic keypoints Keypoint visibility labels Geometrically consistent annotations This makes SASD particularly suitable for keypoint-based pose estimation, learnable PnP frameworks, and multi-task learning. Camera Sampling Strategy Sampling ranges: Azimuth angle: [0°, 360°] Elevation angle: [-75°, 75°] Distance: 100 m – 600 m Spherical spatial distribution of sampled camera poses in the training set. (Images: assets/camera_position_3d_distribution.svg, assets/camera_position_3d_point_cloud(1).svg) Keypoint Definition A set of semantic 3D keypoints is predefined on the spacecraft model: Distributed across the main body and solar panels Includes symmetric structures for ambiguity analysis Defined in the local coordinate system for cross-sample consistency Spatial layout and visibility of predefined keypoints on the FGST spacecraft. Left:Front view; Right:Back view. (Images: assets/4.svg, assets/3.svg) Dataset Statistics Total images: 10,500 Training set: 10,000 Test set: 500 Resolution: 1920 × 1200 Camera parameters (aligned with SPEED dataset): Parameter Description Value Nu Image width (pixels) 1920 Nv Image height (pixels) 1200 fx Focal length (horizontal) 0.0172 m fy Focal length (vertical) 0.0172 m Px Pixel size (horizontal) 5.86 µm Py Pixel size (vertical) 5.86 µm