Deck 12: Raster/Vector Interchangeability

What function converts a Labels element into exact polygon/multipolygon Shapes?

to_polygons()

What function converts a Labels element into approximate circle Shapes?

to_circles()

What's a known limitation of to_circles()?

For a label with multiple disconnected components (e.g. a multipolygon), the resulting circle's centroid can end up outside the actual shape.

What does target_unit_to_pixels control in rasterize()?

The output image resolution → e.g. 0.1 produces roughly ⅒ the pixels.

What does default rasterization do when shapes overlap?

Counts overlaps → pixel values are 0, 1, 2... depending on how many shapes cover that pixel.

What does instance_key_as_default_value_key=True do, and what's its limitation?

Gives each shape a distinct pixel value instead of a count but where shapes overlap, only the first shape is kept.

What does return_single_channel=False do, and when is it recommended?

Produces one channel per shape (no overlap information lost) → recommended only for a small number of large shapes, since memory cost scales with shape count.

What is rasterize_bins() specifically designed for?

Grid-like "binned" data (like Visium HD) with millions of geometries arranged in a regular but possibly rotated grid.

Why should you never call .compute() on the full output of rasterize_bins()?

It's a lazy object that could represent ~180 GB of data for a real Visium HD dataset → access it channel-wise instead (.sel(c=...)).

What extra information does rasterize_bins() compute and store?

The affine transformation needed to align the rasterized (possibly rotated) grid with the original high-resolution image.