Windows — Install Everything at Once¶

← Back to Windows Installation

If you don’t want to think about which pieces you need and just want all of JuSPICE’s features working right away — every image-analysis method, every deep-learning model, every file format reader, plus IOPaint and NASA MicroNet each in their own dedicated environment — run the commands below from PowerShell, in order:

git clone https://jugit.fz-juelich.de/iet-1/juspice.git
cd juspice
uv venv venvs\.venv --python 3.12
venvs\.venv\Scripts\Activate.ps1
uv sync --all-extras --active

$env:UV_PROJECT_ENVIRONMENT = "venvs\.venv_iopaint"
uv sync --python 3.12 --only-group iopaint --no-install-project

$env:UV_PROJECT_ENVIRONMENT = "venvs\.venv_micronet"
uv sync --python 3.12 --extra dl
uv pip install --python venvs\.venv_micronet\Scripts\python.exe `
    git+https://github.com/nasa/pretrained-microscopy-models

Heads up: this downloads a large amount of software — deep-learning libraries such as PyTorch and TensorFlow are several hundred megabytes each, and this sets up three separate environments (venvs\.venv, venvs\.venv_iopaint, venvs\.venv_micronet), so expect this to use around 6-7 GB of disk space for the libraries alone (see the disk-space note below for what each feature additionally downloads the first time you use it) and take a while on a normal internet connection. In exchange, once it finishes you have everything ready to go — no extra installation steps later, no matter which notebook or feature you try. If disk space is limited, or you’d rather understand what you’re installing and add pieces one at a time, see the step-by-step installation instead — it installs a small core first and lets you add exactly the extra features you need.

This command reads a file called uv.lock, which is included in the JuSPICE repository and records the exact version of every package that is known to work together. That means everyone who runs this command gets an identical, reproducible setup, rather than whatever happens to be newest on the day they install it. If you’re contributing to JuSPICE and change pyproject.toml (the file that lists which packages JuSPICE depends on), run uv lock afterwards to keep uv.lock in sync with it.

venvs\.venv_iopaint and venvs\.venv_micronet are kept separate from venvs\.venv because IOPaint and MicroNet each pin their own, incompatible versions of shared packages (diffusers/ huggingface_hub/torch, and segmentation-models-pytorch/ timm respectively) — all three environments’ exact package versions are recorded in this same uv.lock file, without any of them ever getting mixed together.

⚠️ A note on disk space

The 6-7 GB above is just the Python libraries. Several deep-learning features also download their own separate, pretrained model the first time you use them — automatically, with no extra command needed:

SAM-1 checkpoint (spice.segmentation.run_sam1()) — about 2.4 GB
SAM-2 checkpoint (spice.segmentation.run_sam2()) — about 0.9 GB
MicroNet's pretrained encoder (spice.segmentation.run_micronet()) — about 0.1 GB. The example notebook then trains its own small model on top of it, which saves roughly another 1.7 GB of checkpoint files locally — that part is created by training on your computer, not downloaded from anywhere.
Stable Diffusion XL inpainting model (spice.inpaint.stable_diffusion()) — about 6.5 GB
Stable Diffusion v1.5 img2img model (spice.synth_generation.generate(method_name='SDiff')) — about 5.1 GB. This is a different model from the one above — using both features downloads both.
IOPaint's default background-fill model, used by every augmenting synthetic-data method — about 0.2 GB, downloaded into its own separate venvs\.venv_iopaint environment (about 1.2 GB by itself — see IOPaint Installation)

All of these except IOPaint's model are saved into JuSPICE's own Pretrained_models/ folder — see Library Structure. Using every one of these features at least once therefore needs roughly 2.5 GB of libraries plus about 15 GB of downloaded models — call it ~18 GB in total, before counting your own datasets and generated images. (All of these sizes come from one test installation by the developer; your own numbers may differ a little, since library and model versions change over time.)