dgs.utils.state.State

class dgs.utils.state.State(*args, bbox: torchvision.tv_tensors.BoundingBoxes, validate: bool = True, **kwargs)

Class for storing one or multiple samples of data as a ‘State’.

Batch Size

Even if the batch size of a State is 1, or even zero (!), the dimension containing the batch size should always be present.

Validation

By default, this object validates all new inputs. If you validate elsewhere, use an existing dataset, or you don’t want validation for performance reasons, validation can be turned off.

Additional Values

The model might be given additional values during initialization, or at any time using the given setters or the get_item call. Additionally, the object can compute / load further values.

All args and keyword args can be accessed using the States’ properties. Additionally, the underlying dict structure (‘self.data’) can be used, but this does not allow validation nor on the fly computation of additional values. So make sure you do so, if needed.

keypoints (torch.Tensor)

The key points for this bounding box as torch tensor in global coordinates.

Shape [B x J x 2|3]

filepath (FilePaths)

The respective filepath(s) of every image.

Length B.

person_id (torch.Tensor)

The person id, only required for training and validation.

Shape [B].

class_id (torch.Tensor)

The class id, only required for training and validation.

Shape [B].

device (Device)

The torch device to use. If the device is not given, the device of bbox is used as the default.

heatmap (torch.Tensor)

The heatmap of this bounding box. Currently not used.

Shape [B x J x h x w].

image (Images)

A list containing the original image(s) as tv_tensors.Image object.

A list of length B containing images of shape [1 x C x H x W].

image_crop (Image)

The content of the original image cropped using the bbox.

Shape [B x C x h x w]

joint_weight (torch.Tensor)

Some kind of joint- or key-point confidence. E.g., the joint confidence score (JCS) of AlphaPose or the joint visibility of PoseTrack21.

Shape [B x J x 1]

keypoints_local (torch.Tensor)

The key points for this bounding box as torch tensor in local coordinates.

Shape [B x J x 2|3]

param bbox:

One single bounding box as torchvision bounding box in global coordinates.

Shape [B x 4]

type bbox:

tv_tensors.BoundingBoxes

param kwargs:

Additional keyword arguments as shown in the ‘Additional Values’ section.

__init__(*args, bbox: torchvision.tv_tensors.BoundingBoxes, validate: bool = True, **kwargs) → None

Methods

cast_joint_weight([dtype, decimals, overwrite])	Cast and return the joint weight as tensor.
clean([keys])	Given a state, remove one or more keys to free up memory.
clear()
copy()	Obtain a copy of this state.
draw
draw_individually
extract(idx)	Extract the i-th State from a batch B of states.
fromkeys(iterable[, value])
get(k[,d])
items()
keypoints_and_weights_from_paths(paths[, ...])	Given a tuple of paths, load the (local) key-points and weights from these paths.
keys()
load_image([store])	Load the images using the filepaths of this object.
load_image_crop([store])	Load the images crops using the crop_paths of this object.
pop(k[,d])	If key is not found, d is returned if given, otherwise KeyError is raised.
popitem()	as a 2-tuple; but raise KeyError if D is empty.
setdefault(k[,d])
split()	Given a batched State object, split it into a list of single State objects.
to(*args, **kwargs)	Override torch.Tensor.to() for the whole object.
update([E, ]**F)	If E present and has a .keys() method, does: for k in E: D[k] = E[k] If E present and lacks .keys() method, does: for (k, v) in E: D[k] = v In either case, this is followed by: for k, v in F.items(): D[k] = v
values()

Attributes

B	Get the batch size.
J	Get the number of joints in every skeleton.
bbox	Get this States bounding-box.
bbox_relative	Get the bounding box coordinates in relation to the width and height of the full image.
class_id	Get the class-ID of the bounding-boxes.
crop_path	Get the path to the image crops.
device	Get the device of this State.
filepath	If data filepath has a single entry, return the filepath as a string, otherwise return the list.
image	Get the original image(s) of this State.
image_crop	Get the image crop(s) of this State.
joint_dim	Get the dimensionality of the joints.
joint_weight	Get the weight of the joints.
keypoints	Get the key-points.
keypoints_local	Get the local key-points.
keypoints_relative	Get the global key points in coordinates relative to the full image size.
person_id	Get the ID of the respective person shown on the bounding-box.
track_id	Get the ID of the tracks associated with the respective bounding-boxes.
validate	Whether to validate the inputs into this state.
data	All the data in this state as a dict.