Data associated to the publication: A Synthetic Image Dataset for Robust Hazard Detection in Asteroid Environment

This dataset was created as described in the publication: "A Synthetic Image Dataset for Robust Hazard Detection in Asteroid Environment", accepted to the ESA GNC&ICATT Conference 2026. This dataset includes simulated images from the asteroid generated from the work of Caroselli et al. 2024 [1]. The dataset comprises three different partitions: - CLEAN: Asteroid images with boulder segmentation masks - NOMINAL: Asteroid images with rendered solar flares and glares and associated annotation masks - PERTURBED: NOMINAL images with representative perturbations - Sensor Noise (Johnson Noise) - Optics Degradation (Blurriness) - Vignetting The CLEAN dataset is generated via a structured process to guarantee uniform coverage of illumination conditions (Sun Phase Angle) across the whole images. The NOMINAL dataset is generated rendering flares and glares with the Albumentations library [2]. A split of 83%/17% in the images is introduced to balance the Sun position in the sky which is related to straylight rendering. Straylight injection parameters are also engineered to be uniformly covered within their boundaries. The PERTURBED dataset is generated injecting the same proportion of all the possible combinations of considered perturbations, ensuring uniform coverage of the parameters defining them. Such dataset is just for test purposes, therefore no annotation is provided. Images size is 1024,1024,3. Label masks size is 1024,1024,. CLEAN and NOMINAL dataset have 20000 images. PERTURBED dataset has 4000 images. [1] E. Caroselli, F. Belien, A. Falke, F. Curti, and R. Förstner, “Deep learning-based passive hazard detection for asteroid landing in unexplored environment,” in Proceedings of the 44th Annual American Astronautical Society Guidance, Navigation, and Control Conference, 2022, pp. 319–334, Springer, 2022. [2] A. Buslaev, V. I. Iglovikov, E. Khvedchenya, A. Parinov, M. Druzhinin, and A. A. Kalinin, “Albumentations: Fast and flexible image augmentations,” Information, vol. 11, no. 2, 2020.