Tiling Strategies¶

GlassCut provides a grid-based tiler out of the box, but the Tiler abstract base class is designed to be extended with custom strategies.

Important

The built-in GridTiler is provided as a reference implementation to demonstrate the tiling API and common patterns. For production use, we strongly recommend implementing your own tiling strategy tailored to your specific slide dimensions, tissue types, and downstream task requirements.

GridTiler¶

The GridTiler extracts tiles on a regular grid with configurable overlap and tissue-prescreening.

from glasscut import Slide, GridTiler

tiler = GridTiler(
    tile_size=(512, 512),
    magnification=20
)

slide = Slide("slide.svs")

for tile in tiler.extract(slide):
    tile.save(f"tiles/{tile.coords}.png")

Grid tiling visualization — Grid tiles overlaid on heart tissue slide; green tiles pass tissue prescreening¶

Key Parameters¶

Parameter	Default	Description
`tile_size`	`(512, 512)`	Dimensions of each tile in pixels
`magnification`	`20`	Target magnification level
`overlap`	`0`	Pixel overlap between adjacent tiles
`min_tissue_ratio`	`0.2`	Minimum tissue fraction to keep a tile (0–1)
`tissue_detector`	`None`	A `TissueDetector` instance
`transforms`	`None`	List of `TileTransform` callables

Performance¶

GridTiler uses batched parallel extraction via ThreadPoolExecutor. You can control parallelism with n_workers and batch_size parameters in the extract() method.

Visualisation¶

viz = tiler.visualize(slide)
viz.save("grid_overview.png")

Custom Tilers¶

Implement a custom tiler by subclassing Tiler:

from glasscut.tiler import Tiler
from glasscut.slides import Slide
from glasscut.tile import Tile

class MyCustomTiler(Tiler):
    def get_tile_boxes(self, slide: Slide):
        return [(0, 0, 512, 512)]

    def extract(self, slide, *, n_workers=4, batch_size=128):
        for box in self.get_tile_boxes(slide):
            tile = slide.extract_tile(box[:2], box[2:], magnification=20)
            yield tile