Prediction API¶
run_instance_segmentation(image, net_size=256, mode='2D', isotrope=True, patches=True, voxel_size=(1.0, 1.0, 1.0), z_axis=0, weights_path=None)
¶
Segment a grayscale image into labeled cell instances.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
image
|
ndarray
|
Two-dimensional image or three-dimensional volume. |
required |
net_size
|
int
|
Published model patch size: 256 for 2D; 128 or 256 for 3D. |
256
|
mode
|
('2D', '3D')
|
Dimension of the input image. |
"2D"
|
isotrope
|
bool
|
Rescale a 3D volume to isotropic spacing before inference. |
True
|
patches
|
bool
|
Predict on overlapping patches. If false, resize the whole image to the model input size before prediction. |
True
|
voxel_size
|
tuple of float
|
Physical spacing for each array axis. Used for 3D isotropic rescaling. |
(1.0, 1.0, 1.0)
|
z_axis
|
int
|
Axis treated as the axial direction for 3D rescaling. |
0
|
weights_path
|
path - like
|
Local model weights. If omitted, use the published model cache. |
None
|
Returns:
| Type | Description |
|---|---|
ndarray
|
Integer instance labels, with the same shape as |
Raises:
| Type | Description |
|---|---|
ValueError
|
If the dimension, patch size, or voxel spacing is invalid. |
FileNotFoundError
|
If |
run_instance_segmentation is the supported Python prediction entry point.
It returns an integer NumPy array with the same shape as the input image.