nv-segment-ct

To generate a structured JSON evidence payload for CT segmentation, you can use an engineering wrapper that delegates inference to the HuggingFace VISTA3D pipeline, processes your CT volume, and outputs a reproducible label map alongside the report.

Yes, you need a CUDA GPU to run HuggingFace VISTA3D CT segmentation using this wrapper, because the underlying PyTorch model bundle requires CUDA-compatible hardware for inference operations.

The wrapper handles the 832 MB VISTA3D model bundle download automatically using HuggingFacePipelineHelper, placing the assets into a local bundle directory and caching fixture data for tests on demand.

No, you cannot use this CT segmentation wrapper for clinical diagnosis; it is strictly designed for engineering verification, QA, and research scenarios where reproducible JSON reports are required.

The VISTA3D CT segmentation wrapper expects a CT volume as input and produces a label map and a structured result JSON file as outputs, encapsulating input, output, and geometry details for verification.

Running CT volume segmentation to output a JSON report requires numpy and nibabel for volume processing, torch for GPU inference, and typer for the command-line script interface.