System Requirements

This page documents the supported architectures and recommended minimum hardware for NRTK in both its forms: the nrtk Python library and the nrtk-perturber container. Requirements are largely the same for both forms and are described together, with differences called out explicitly.

Note

Hardware figures below are rough order-of-magnitude estimates intended to cover the minimum to function plus a recommended configuration for comfortable use. Actual needs depend on image size, dataset size, the number of perturbation parameter combinations, and which perturbers you use.

Supported Architectures

Python library (nrtk)

• Python: CPython 3.10 – 3.14.

• Operating systems: Linux, macOS, and Windows via WSL. (NRTK is tested on Unix-based systems; native Windows is not supported.)

Container (nrtk-perturber)

• Architecture: linux/amd64 (x86_64) only.

• Bundled Python: the image is built on an official python base image and installs all optional extras.

Recommended Minimum Hardware

Important

Memory is the limiting factor on dataset size for batch perturbation. nrtk-perturber accumulates the entire perturbed output dataset(s) in memory before writing them to disk, so peak RAM scales roughly with (decoded image bytes in the dataset) × (number of perturbation parameter combinations). For large datasets or many parameter combinations, partition the dataset or reduce the number of combinations to stay within available RAM.

Resource	Library, classical perturbers only	Library, with diffusion	Container	Notes
CPU	2 cores min, 4+ recommended	4–8 cores recommended	4–8 cores recommended	pyBSM and water-droplet perturbers use Numba JIT and scale with cores; CPU diffusion benefits strongly from more cores.
GPU	None	Optional NVIDIA CUDA, ≥ 6–8 GB VRAM	None	Container is CPU-only. Diffusion perturber supports CPU fallback but is slow.
Memory (RAM)	2 GB min, 4 GB recommended	8–16 GB recommended	8–16 GB recommended	The diffusion model loads fully into memory (~a few GB). See the data-size note below.
Storage	~0.5 GB install	~3–6 GB (PyTorch + diffusers + model cache)	~3–6 GB (PyTorch + diffusers + model cache)	First diffusion use downloads timbrooks/instruct-pix2pix (~a few GB) to the Hugging Face cache (~/.cache/huggingface). The container image itself is several GB.