Mask¶
The aloscene Mask object represents a binary or float mask. It can be used to represent different objects,
for example occlusions mask for flow and disparity use cases.
Note
The values of a Mask are between 0. and 1., to allow to partially/completely mask another tensor
by multiplying it by the mask.
Basic Use¶
A Mask object can be initialized from a path to a mask file:
from aloscene import Mask
mask = Mask("path/to/mask.png")
or from an existing tensor:
import torch
mask_float_tensor = torch.rand((1,400,600))
mask = Mask(mask_tensor, names=("C","H","W"))
Mask API¶
- class aloscene.mask.Mask(x, labels: Optional[Union[dict, aloscene.labels.Labels]] = None, *args, **kwargs)¶
Bases:
aloscene.tensors.spatial_augmented_tensor.SpatialAugmentedTensorBinary or Float Mask
- Parameters
- x :
path to the mask file (png) or tensor (values between 0. and 1.)
- append_labels(labels, name=None)¶
Attach a set of labels to the masks.
- Parameters
- labels: aloscene.Labels
Set of labels to attached to the masks
- name: str
If none, the label will be attached without name (if possible). Otherwise if no other unnamed labels are attached to the frame, the labels will be added to the set of labels.
- get_view(frame=None, size=None, labels_set=None, color_by_cat=False, **kwargs)¶
Get view of segmentation mask and used it in a input Frame
- Parameters
- frameTensor, optional
Frame where the segmentation mask will be displayed, by default None
- sizetuple, optional
Size of a desired masks, by default not-resize
- labels_setstr, optional
Set of labels to show in segmentation when multiple labels are defined, by default None
- color_by_catbool, optional
Set same color by category ID, by default False
- Returns
- Renderer.View
Frame view, ready to render
- Raises
- Exception
Input frame must be a aloscene.Frame object
- iou_with(mask2)¶
IoU calculation between mask2 and itself
- Parameters
- mask2aloscene.Mask
Masks with size (M,H,W)
- Returns
- torch.Tensor
IoU matrix of size (N,M)
- rtype
Tensor..
- mask2id(labels_set=None, return_ann=False, return_cats=False)¶
Create a panoptic view of the frame, where each pixel represent one class
- Parameters
- labels_setstr, optional
If multilabels are handled, get mask_id by a set of label desired, by default None
- return_annbool, optional
Return annotations to get_view function, by default False
- return_catsbool, optional
Return categories ID instance ID, by default False.
- Returns
- np.array
Array of (H,W) dimensions, where each value represent one class