VISTA3D

English简体中文

A unified 3D foundation model for automatic and interactive medical segmentation. The paper's research implementation is the MONAI 1.3 VISTA3D codebase.

Official code Paper Weights

← Model atlas

At a glance

Field Details
Model type 3D automatic and point-prompt segmentation
Research team / organization Project MONAI / NVIDIA
Input / modality The CVPR 2025 research checkpoint is CT-focused and supports 127 anatomy/lesion classes.
Prompt / interaction Class prompts for automatic segmentation; positive/negative 3D point clicks for interactive segmentation and refinement.
Code license Code: Apache-2.0. The research checkpoint is distributed under NVIDIA’s open model terms; check the linked model card. VISTA3D and CTMR variants differ in modality, checkpoint and task.

📚 Paper and publication

VISTA3D: A Unified Segmentation Foundation Model For 3D Medical Imaging — IEEE/CVF CVPR 2025.

📦 Checkpoints and access

The research code automatically fetches nvidia/NV-Segment-CT/vista3d_pretrained_model/model_monai1.3.pt. MONAI 1.4+ bundles use different checkpoint naming. NV-Segment-CTMR is a separate automatic CT/MR model.

The code, model checkpoint and any required foundation-model checkpoint can have different terms. Review each linked source before use or redistribution.

⚙️ Official setup and inference

1
2
3
4
5
6
7
8
9
git clone https://github.com/Project-MONAI/VISTA.git
cd VISTA/vista3d
conda create -n vista3d python=3.9 -y
conda activate vista3d
pip install -r requirements.txt
# Automatic organ segmentation (label 1 is liver in the official mapping):
python -m scripts.infer --config_file 'configs/infer.yaml' - infer \
--image_file example-1.nii.gz --label_prompt '[1]' --save_mask true
# Interactive clicks use --point and --point_label; see the official guide.

⚠️ Version and practical notes

The research repository is CT-focused and distinct from the newer CTMR automatic model. State whether inference uses class prompts, points, or automatic segmentation followed by point refinement.

📜 License

Code: Apache-2.0. The research checkpoint is distributed under NVIDIA’s open model terms; check the linked model card. VISTA3D and CTMR variants differ in modality, checkpoint and task.

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.