---

name: data-io description: "Load and save images with skimage.data and skimage.io, reason about NumPy image conventions and dtype conversion, and handle safe local round-trips and collections." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Data I/O

Use this sub-skill when the task is about bundled sample images, local image files, image collections, or the conversion boundary between file data and NumPy arrays.

Route away from this sub-skill when the task is mainly about:

• color, exposure, filtering, or restoration: use ../enhancement/SKILL.md
• feature extraction, measurements, or image metrics: use ../analysis/SKILL.md
• masks, morphology, watershed, or synthetic shapes: use ../segmentation-and-shapes/SKILL.md
• geometric transforms, warps, or registration: use ../transform-registration/SKILL.md
• video decoding or plugin internals as the primary task: keep that in references/troubleshooting.md and prefer imageio or another direct video library instead.

Start Here

• Read references/workflows.md for bundled sample data, safe local I/O, dtype conversion, and NumPy image conventions.
• Read references/troubleshooting.md when imports fail, astype changes the image unexpectedly, skimage.io plugin warnings appear, or collection and download behavior is confusing.
• Run scripts/check_install.py for a safe import and temp-file round-trip smoke test.

Fast Paths

• Use skimage.data.camera(), coins(), astronaut(), text(), logo(), horse(), or binary_blobs() for bundled inputs.
• Load local files with skimage.io.imread(path); write temporary round trips with skimage.io.imsave(path, image).
• Convert data with skimage.util.img_as_float, img_as_ubyte, img_as_uint, img_as_int, or img_as_bool; never use astype to rescale an image.
• Remember the core image convention: rows and columns come first, channels are explicit, and channel_axis is how most multichannel APIs stay clear.
• Use imread_collection when you want the convenience wrapper around ImageCollection; use ImageCollection directly for file sequences or custom loaders, and use MultiImage for multi-frame TIFFs when a single file should expose a frame list.
• Use view_as_windows, invert, montage, and lookfor as lightweight utilities in this route, but keep them secondary to data loading and saving.
• Prefer imageio or another direct I/O package when the task needs modern plugin control, video handling, or format-specific features. skimage.io plugin arguments are deprecated.
• Optional offline bulk download is a setup step, not a primary route; see troubleshooting before using skimage.data.download_all().

Reference Map

• references/workflows.md — sample-data, image I/O, dtype, collections, and NumPy conventions.
• references/troubleshooting.md — install, deprecation, download, collection, and dtype pitfalls.
• scripts/check_install.py — safe import and local image round-trip smoke check.

Smoke Check

Run from the sub-skill root:

python scripts/check_install.py