AKM Dataset: 10,000 Synthetic Images of a Textureless Space Motor

A dataset of 10,000 synthetic images of the Apogee Kick Motor (AKM), a radially symmetric and textureless space object. Images are 1024x1024 pixels, generated with Blender under orbital lighting conditions, with target positions and orientations sampled from uniform random distributions. The dataset was created by researchers at the University of Luxembourg and presented at the 73rd International Astronautical Congress in 2022.

Use Cases

• Training pose estimation models based on synthetic images of a textureless object.
• Benchmarking neural network performance on space object recognition tasks using the provided ground truth pose labels.
• Simulating and testing computer vision algorithms for satellite servicing or space situational awareness based on the orbital lighting and camera model.
• Developing domain adaptation techniques for synthetic-to-real transfer in space imagery.

Strengths

• Contains 10,000 high-resolution images (1024 x 1024 pixels).
• Includes precise ground truth pose labels (translation and quaternion orientation) for each image.
• Synthetic generation allows for controlled variation in target distance (1-10 meters) and orientation.

Limitations

• Data is purely synthetic, which may not fully capture the noise and artifacts of real orbital imagery.
• Column-level documentation is absent; field semantics must be inferred from the description after download.
• Row count for the accompanying ground truth CSV is unknown, which may limit suitability assessment.

Provenance

Source

University of Luxembourg

Collection Method

Synthetically generated using Blender software under simulated orbital lighting conditions.