lacss.modules
lacss.modules.Lacss
Bases: Module, DefaultUnpicklerMixin
Main class for LACSS model
Attributes:
| Name | Type | Description |
|---|---|---|
backbone |
Module
|
CNN backbone |
detector |
Module
|
detection head to predict cell locations |
segmentor |
Module | None
|
The segmentation head |
__call__(image, *, image_mask=None, video_refs=None)
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
image
|
ArrayLike
|
[H, W, C] or [D, H, W, C] |
required |
Returns: a dict of model outputs
lacss.modules.LPN
Bases: Module, DefaultUnpicklerMixin
Location detection head
Attributes:
| Name | Type | Description |
|---|---|---|
nms_threshold |
float
|
non-max-supression threshold, if performing nms on detected locations. |
pre_nms_topk |
int
|
max number of detections to be processed regardless of nms, ignored if negative |
max_output |
int
|
number of detection outputs |
lacss.modules.Segmentor
Bases: Module, DefaultUnpicklerMixin
LACSS segmentation head.
Attributes:
| Name | Type | Description |
|---|---|---|
n_cls |
num of classes |
|
feature_scale |
int
|
the spatail scale of the feature level |
instance_crop_size |
int
|
Crop size for segmentation. |
pos_emb_shape |
Sequence[int]
|
Dim of the learned position encoder. |
__call__(feature, locations)
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
features
|
[H, W, C] features from the backbone. |
required | |
locations
|
ArrayLike
|
[N, 2] |
required |
Returns:
| Type | Description |
|---|---|
dict
|
A nested dictionary of values representing segmentation outputs. * segmentor/logits: all segment logits * predictions/segmentations: logits of designated class [N, D, ps, ps] * predictions/segmentation_y0_coord: y coord of patch top-left corner [N] * predictions/segmentation_x0_coord: x coord of patch top-left corner [N] * predictions/segmentation_is_valid: mask for valid patches [N] |