merge_batch_to_canvas

tiatoolbox . models . engine . semantic_segmentor . merge_batch_to_canvas

merge_batch_to_canvas(blocks, output_locations, merged_shape)

Merge patch-level predictions into a single canvas.

This function aggregates overlapping patch predictions into a unified output canvas and maintains a count map to normalize overlapping regions.

Parameters:

• blocks (np.ndarray) – Array of predicted blocks with shape (N, H, W, C), where N is the number of patches.

• output_locations (np.ndarray) – Array of coordinates for each block in the format [start_x, start_y, end_x, end_y] with shape (N, 4).

• merged_shape (tuple[int, int, int]) – Shape of the final merged canvas (H, W, C).

Returns:

• canvas: Merged prediction map of shape (H, W, C).

• count: Count map indicating how many times each pixel was updated, shape (H, W).

Return type:

tuple[np.ndarray, np.ndarray]