Robustness Analysis of Data-Driven Image Processing Algorithms Applied to the Hera Mission

This database accompanies the article titled "Robustness Analysis of Data-Driven Image Processing Algorithms Applied to the Hera Mission" by Kaluthantrige A. , Pugliatti M. , Jinglang F. , Topputo F. , Gil-Fernandez J. Journal of Spacecraft and Rockets (2025).

The dataset is structured as follows according to the resolution of the images (square 1024px or 128px images): 

• 1024x1024 (~27 GB): it contains image-label pairs for the original 1024x1024 renderings of the Didymos system with artificial noise added. This dataset is used to train M1 in [1]. The images are saved as .png files, while the labels are saved as .mat and contain information such as: 

  • CoM: the projection of the center of mass of Didymos in the image plane (px)

  • range: the distance between the camera and the CoM of the body (km)

  • pa: the phase angle, the angle between sun-body-spacecraft (deg)

• 128x128 (~0.5 GB): it contains image-label pairs for the processed 128x128 images of the Didymos system with artificial noise added. This dataset is used to train M2 in [1]. This set is a lightweight, processed version of the previous one, specifically designed for training neural networks without further pre-processing. The images of this dataset are cropped and resized versions of the ones in the 1024x1024 folder. It should serve as an example of a pre-processed set of the original one and represent the exact one used by the authors to train M2. The images are saved as tensors in .mat files. The labels are also saved as .mat files and contain information such as: 

  • CoM: the projection of the center of mass of Didymos in the image plane (px). This refers to the 128x128 images, but can be scaled to the original 1024x1024 images. 

  • range: the scaled distance between the camera and the CoM of the body (km). This refers to the 128x128 images, but can be scaled to the original 1024x1024 images. 

  • pa: the phase angle, the angle between sun-body-spacecraft (deg). 

  • selected_target_size: the size of the minimum square bounding box that contains the body (px). This refers to the size of the body in the original 1024x1024 images.

• DB_table: an image containing a table summarizing the main properties of the different datasets. For more details, please see [1]. 

Within each subfolder, there is a preview image of the dataset itself with a random selection of 50 images. The images also contain the CoM projection (represented as a red cross) and a text label at the top-left corner with the pa and range (for the 1024x1024 images) or the pa, range, and selected_target_size (for the 128x128 images). Note that in the 128x128 set, to re-scale the range to a quantity of physical interest, you need to multiply it 128 and divide it by selected_target_size. This should return the physical range between values of 10km and 40km from Didymos.

For more details on the datasets and how they are used, please see [1]. The dataset is created using CORTO, an open-source rendering tool, see [2,3] for more details. If you encounter an issue with the dataset, don't hesitate to drop a line by email! 

References:

[1] "Robustness Analysis of Data-Driven Image Processing Algorithms Applied to the Hera Mission" by Kaluthantrige A. , Pugliatti M. , Jinglang F. , Topputo F. , Gil-Fernandez J. Journal of Spacecraft and Rockets (2025).

[2] "CORTO: The Celestial Object Rendering TOol at DART Lab", by Pugliatti M., Buonagura C. , and Topputo F. Sensors. Volume 3, pp9595 (2023). doi: 10.3390/s23239595

[3] CORTO GitHub repository: https://github.com/MattiaPugliatti/corto