lazyslide.seg.semantic

semantic(wsi, model, tile_key='tiles', class_names=None, transform=None, mode='gaussian', sigma_scale=0.125, low_memory=False, threshold=0.5, ignore_index=0, buffer_px=2, chunk_size=512, batch_size=4, num_workers=0, device=None, amp=None, autocast_dtype=None, pbar=None, key_added='anatomical_structures')

Semantic segmentation for the whole slide image.

Parameters:

wsiWSIData

The WSIData object to work on.

modelSegmentationModel

The segmentation model.

tile_keystr, default: “tiles”

The key of the tile table.

class_nameslist of str or dict, optional

The class names for the segmentation. Either a list of class names or a dict mapping class indices to names.

transformcallable, default: None

The transformation to apply to each input tile before segmentation.

mode{“constant”, “gaussian”}, default: “gaussian”

The probability distribution to apply for the prediction map. “constant” uses uniform weights, “gaussian” applies a Gaussian weighting.

sigma_scalefloat, default: 0.125

The scale of the Gaussian sigma for the importance map if mode is “gaussian”.

low_memorybool, default: False

Whether to use a low-memory mode for processing large slides.

thresholdfloat, default: 0.5

The threshold to binarize the probability map for segmentation.

ignore_indexint or None, default: 0

The index to ignore during segmentation (e.g., background).

buffer_pxint, default: 2

The buffer in pixels to apply when merging polygons.

chunk_sizeint, default: 512

The size of chunks to process at a time when merging probability maps.

batch_sizeint, default: 4

The batch size for segmentation.

num_workersint, default: 0

The number of workers for data loading.

devicestr, default: None

The device for the model (e.g., “cpu” or “cuda”). If None, automatically selected.

ampbool, optional, default: False

Whether to use automatic mixed precision.

autocast_dtypetorch.dtype, optional, default: torch.float16

The dtype for automatic mixed precision.

pbarbool, default: True

Whether to show the progress bar.

key_addedstr, default: “anatomical_structures”

The key for the added instance shapes in the WSIData object.

Returns:

None

The segmentation results are added to the WSIData object under the specified key.