v1.0.0#

This is the first stable release of the Natural Robustness Toolkit (NRTK). The API is hardened and will follow semantic versioning going forward: patch releases (1.0.x) for bug fixes, minor releases (1.x.0) for backward-compatible features, and major releases (x.0.0) for breaking changes.

v1.0.0 establishes the full foundation of NRTK for still-imagery computer vision robustness evaluation. Future releases will expand into full-motion video, additional sensor modalities, and broader operational domains.

There are no functional changes from v0.27.2 — this is a drop-in replacement.

What’s in v1.0.0#

27 perturber implementations across five categories:

  • Photometric (12) — blur (average, Gaussian, median), enhancement (brightness, color, contrast, sharpness), and noise (Gaussian, salt, pepper, salt-and-pepper, speckle)

  • Optical (8) — physics-based perturbations via pyBSM (pybsm, defocus, jitter, turbulence aperture, detector, circular aperture) and radial distortion

  • Geometric (4) — random crop, translation, rotation, and scale

  • Environment (2) — haze and water droplets

  • Generative (experimental) (1) — diffusion-based perturbation

  • Plus ComposePerturber for chaining multiple perturbations and AlbumentationsPerturber for integration with the Albumentations library

4 perturbation factories for systematic parameter sweeps: step, linspace, one-step, and multivariate.

8 operational risk modules mapping real-world degradation factors to NRTK perturbations (atmospheric turbulence, defocus, extreme illumination, haze, high-frequency vibration, radial distortion, sensor noise and resolution, water droplets), with an interactive risk matrix.

JATIC ecosystem integration — NRTK is part of the JATIC product suite for AI Test & Evaluation. The nrtk.interop module provides full MAITE compliance, enabling NRTK perturbations to be applied directly to MAITE-wrapped datasets and models alongside other JATIC tools such as DataEval and XAITK-Saliency. This includes augmentation adapters for both classification and detection tasks.

Validation and Trust page documenting current validation status, known limitations, and a quarterly-updated roadmap.

Comprehensive documentation following Diátaxis structure (Getting Started, Tutorials, How-To Guides, Explanations, Reference) with 19 example notebooks spanning core tutorials, MAITE workflows, and XAITK-Saliency integration.

Modular dependency architecture — NRTK’s core has a minimal dependency footprint, with 10 optional extras that allow users to install only what they need. This makes NRTK deployable into a wide range of environments, including hardened or restricted systems with strict dependency policies. Python 3.10–3.13 supported.

Acknowledgment#

This material is based upon work supported by the Chief Digital and Artificial Intelligence Office under Contract No. 519TC-23-9-2032. The views and conclusions contained herein are those of the author(s) and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the U.S. Government.