Multi-purpose Sequential Image Simulation Dataset of Space Targets

To address the prohibitive acquisition costs of real-world spacecraft imagery and the limitations of existing datasets, we introduces a multi-purpose sequential image dataset for space targets. The dataset comprises two complementary subsets:The first subset consists of sequential simulated images accompanied by fine-grained labels. By incorporating temporal continuity, this sequential data enables computer vision algorithms to learn the dynamic features of on-orbit target motion. Furthermore, the provided fine-grained annotations supply detailed ground truth, establishing a robust foundation for high-precision target detection and local feature extraction.The second subset features high-fidelity sequential images paired with depth maps, precise segmentation masks, and pose annotations. To effectively bridge the sim-to-real gap, these images are synthesized to closely mimic complex space illumination conditions. The inclusion of depth maps captures accurate 3D geometric structures of the space targets. Equipped with this depth modality alongside pixel-level masks and 6D pose labels, this subset serves as an ideal testbed not only for evaluating the robustness of instance segmentation and pose estimation algorithms in highly photorealistic scenarios, but also for supporting advanced research in multi-modal perception and 3D reconstruction tasks.