Glossary#

augmentation#: See perturbation.
COCO#: Common Objects in Context (COCO) — A large-scale dataset that is a widely used benchmark for object detection and segmentation. The COCO format is also used in the JATIC Interoperability module.
environment perturber#: Environment perturbers simulate atmospheric and weather-related effects that occur in real-world imaging conditions.
generative perturber#: Generative perturbers use AI models to transform images through learned representations.
geometric perturber#: Geometric perturbers alter the spatial positioning and orientation of images through transformations such as rotation, scaling, cropping, and translation.
item response curve#: A graphical representation of how mean image scores change based on perturber values.
MAITE#: Modular AI Trustworthy Engineering — a framework for evaluating the trustworthiness and robustness of AI systems using standardized metrics and workflows. View MAITE documentation or NRTK integration.
natural robustness#: A model’s ability to maintain performance despite variations or changes in the environment or inputs that are naturally occurring, not specifically designed for testing or manipulation.
optical perturber#: Optical perturbers simulate physics-based sensor and optical effects.
Optical Transfer Function (OTF)#: A mathematical model that describes how an imaging system reduces detail and sharpness in an image due to physical limitations such as diffraction, motion, or sensor imperfections. In NRTK, perturbing OTF parameters simulates various sensor and environmental effects.
perturbation#: A modification applied to input data to simulate noise, environmental degradation, or sensor artifacts.
perturber#: A reusable component that defines and applies a specific type of perturbation (e.g. haze or blur) to image data.
perturber factory#: A factory method implementation for creating perturbers. Perturbers can be customized by changing thetas and theta keys.
photometric perturber#: Photometric perturbers modify the visual appearance of images by adjusting color, brightness, contrast, sharpness, blur, and noise properties.
pyBSM#: Python Based Sensor Model (pyBSM) — A toolset for modeling and simulating physical sensor effects such as blur, sensor noise, and environmental conditions. View on GitHub.
saliency#: A measure of how much influence a part of an input has on a model’s output.
sensor transformation#: A change applied to image data to simulate different sensor behaviors (e.g. wavelength response, resolution, distortion).
theta key(s)#: A string (or list of strings) that is the name of the pertubrer parameter to modify. A single string is used for generic factories and a list of strings is used for the multivariate factory.
thetas#: A list of values (or list of lists) containing the values for the perturber parameter. A single list is used for generic factories and a list of lists is used for the multivariate factory.
utility perturber#: Utility perturbers enable composition of multiple perturbations or provide integration with third-party augmentation libraries.