๐ฑ๏ธ nnInteractive
A 3D open-set interactive segmentation system trained across diverse volumetric datasets and designed for practical annotation workflows.
At a glanceโ
| Field | Details |
|---|---|
| Model type | Open-set 3D interactive segmentation |
| Research team / organization | German Cancer Research Center (DKFZ) / Helmholtz Imaging |
| Input / modality | 3D CT, MRI, PET and 3D microscopy, among other volumetric images. |
| Prompt / interaction | 2D interactions produce 3D masks; supports points, scribbles, bounding boxes and lasso. |
| Code license | Code: Apache-2.0. Official v1 checkpoint: CC BY-NC-SA 4.0; model terms are more restrictive than the code license. |
๐ Paper and publicationโ
nnInteractive: Redefining 3D Promptable Segmentation โ arXiv:2503.08373 (2025 preprint). The project links the paper and its Python backend.
๐ฆ Checkpoints and accessโ
The official model registry downloads nnInteractive_v1.0 on demand from Hugging Face. It is cached locally under ~/.nninteractive by default.
The code, model checkpoint and any required foundation-model checkpoint can have different terms. Review each linked source before use or redistribution.
โ๏ธ Official setup and inferenceโ
# Install a PyTorch build matching the local CUDA driver, then:
pip install nnInteractive
nninteractive-download-model nnInteractive_v1.0
Use the official minimal inference example with nnInteractiveInferenceSession. A remote GPU server is also available; CPU-only clients can install nninteractive-client and connect to that server.
โ ๏ธ Version and practical notesโ
The official model is distributed as a pretrained checkpoint. Its code and weights have different licenses, so review both before reuse or redistribution. For an agentic workflow built on nnInteractive, see nnInteractive Agent; it adds an MCP/agent layer rather than a new segmentation checkpoint.
๐ Licenseโ
Code: Apache-2.0. Official v1 checkpoint: CC BY-NC-SA 4.0; model terms are more restrictive than the code license.
Citation: Cite the paper above and the upstream project as directed by its authors. This repository links to the original paper; it does not redistribute the code or weights.
Source pages checked: 27 October 2026. Upstream environments, weights and access policies may change; use the linked project as the source of truth.