Download the data#

Download the 42.8 MB teaching ZIP before class. It keeps all 30 time frames while reducing each spatial dimension by a factor of four; the extracted image arrays contain about 122.5 MiB. That reduction lowers download, memory, and rendering demands, but it permanently removes fine spatial detail.

The teaching derivative comes from napari Workshop Datasets: Deciphering the Living Cell through Microscopy and Computation (2026) and is licensed CC BY 4.0. The derivative is archived separately at Zenodo DOI 10.5281/zenodo.22696913. Zenodo is the default download source; GitHub retains an identical mirror.

Recommended: verified command-line download#

Open PowerShell on Windows or Terminal on macOS. Move into the extracted workshop folder—the folder containing pyproject.toml—and preview the download:

uv run scripts/download_data.py --dataset teaching --dry-run

For this script command, uv run uses the script’s own declared and locked environment. It does not change your Jupyter kernel. The script selects the named teaching package, and --dry-run reports what would be downloaded without writing it. The archive is cached in data/downloads and then extracted into data/teaching. Check that the reported size is practical for your connection.

Start the download:

uv run scripts/download_data.py --dataset teaching

The downloader verifies the archive against the checksum recorded in the project manifest before extraction. Rerunning the same command is safe: a previously verified package can be reused, including one downloaded earlier from GitHub. Its bytes and checksum have not changed.

At an ideal sustained 10 Mbit/s the 42.8 MB transfer takes about 35 seconds, or about 3.5 seconds at 100 Mbit/s. Browser, server, Wi-Fi, verification, extraction, and a room full of shared connections add overhead, so these calculations are not a download-time promise.

Success produces this layout:

data/teaching/
├── metadata.json
├── detections.csv
├── tracks.csv
├── ch0/
│   ├── frame_000000.tif
│   └── ...
└── ch1/
    ├── frame_000000.tif
    └── ...

Both channel folders should contain frames numbered from frame_000000.tif through frame_000029.tif. Do not rename individual frames.

Browser download#

If the command-line route is blocked, download teaching.zip from Zenodo in a browser. An identical GitHub mirror is available if Zenodo is unreachable.

Wait for the browser to finish downloading. A partial browser download is not usable.
Open the workshop folder, then open its data folder.
Extract the ZIP there. On Windows, right-click it and choose Extract All…. On macOS, double-click it.
The result must be data/teaching/metadata.json, not data/teaching/teaching/metadata.json. Move the inner folder up one level if extraction created an extra nested folder.
Keep the ZIP until the setup notebook confirms the data can be read.

The scripted route checks file identity automatically. With a browser download, use the setup notebook to check the extracted layout before class.

Five-minute classroom limit#

During class, stop a teaching-package download after five minutes if it is not complete. Tell the instructor and use the supplied USB/local copy, pair with a participant who has the data, or follow the static 2D route. Waiting longer consumes the lesson without teaching napari. This limit is a classroom decision, not a promised download time.

Optional source data#

The full source images are much larger. Preview every source request first:

uv run scripts/download_data.py --dataset source --timepoints 5 --dry-run

The five-frame option downloads the first five raw volumes from both channels plus the tables, about 1.35 GB:

uv run scripts/download_data.py --dataset source --timepoints 5

All 30 source frames require about 8.1 GB to download, before environment and output space:

uv run scripts/download_data.py --dataset source --timepoints 30 --dry-run
uv run scripts/download_data.py --dataset source --timepoints 30

Use the 30-frame command only as optional take-home work on a suitable connection. The complete source image arrays alone occupy about 7.53 GiB in memory if loaded eagerly; the large-data notebook therefore demonstrates lazy loading. Lazy loading does not reduce the download, guarantee fast scrubbing, or remove GPU limits.

An instructor with verified original files can avoid another network transfer:

uv run scripts/download_data.py --dataset source --timepoints 5 --local-source "/path/to/original files"

The quoted path may contain spaces. The script checks the local source before reuse.

What is already computed#

detections.csv contains supplied rows with columns t,z,y,x,amp,int. tracks.csv contains headerless rows in id,t,z,y,x order and represents 2,058 supplied tracks. The workshop reads and displays these tables. Downloading or opening them does not recompute detections or tracks, and no workshop step asks you to run detection or tracking.

The layer names DNA and mitochondria come from the source files. More specific particle identity, voxel calibration, and acquisition timing are awaiting confirmation. Treat t as a frame index and z,y,x as source-voxel coordinates; do not add seconds, micrometres, or a physical scale bar.

Download the data

Contents