davisinteractive.dataset
Davis
Davis(self, davis_root=None)
This class only needs to have the root path specified. Some atributes can be accessible like the sequence list for every subset, or specific information for every sequence like the number of frames, the number of objects or the image size for every sequence. For more information about the sequence attributes available, check this file.
Arguments
- davis_root: String. Path to the DAVIS dataset. This argument can
be left as
None
and specify it as an environtmental variableDATASET_DAVIS
. This usage is useful in the case a group of people is working with the same code and every one has a different path where the DAVIS dataset is stored. The folder name where all DAVIS dataset is stored must be namesDAVIS
.
Attributes
ANNOTATIONS_SUBDIR
: Relative path with respect to the root path where the ground truth masks are stored. (Annotations)SCRIBBLES_SUBDIR
: Relative path with respect to the root path where the scribbles are stored. (Scribbles)RESOLUTION
: Resolution of the dataset used to perform all the evaluation. (480p)sets
: Dictionary. The keys are all the DAVIS dataset subsets and the values are the list of sequences belonging to that subset.dataset
: Dictionary. Contains all the information for the entire dataset. The key is the sequence name and the value is a dictionary of informations such as number of frames, number of objects, etc.years
: List. List with all the versions available from the dataset.
Raises
ValueError
: if neitherdavis_root
or environmental variableDATASET_DAVIS
are specified.
check_files
Davis.check_files(sequences)
Check if all the annotations and scribbles files, required to do the
evaluation are found on davis_root
.
If the scribbles files are not found, it downloads them from the
internet.
Arguments
- sequences: List. List of sequences you want to check.
Raises
FileNotFoundError
: if any required files is not found.
load_images
Davis.load_images(sequence, dtype=uint8)
Arguments
- sequence: String. Sequence name.
- dtype: Numpy Data Type. Data type to return the images. Default
value is
np.uint8
.
Returns
Numpy Array
: Array with all images of the given sequence. The shape
of the array will be (nb_frames x H x W x 3)
and the value
will be the pixel's value with range
: [0, 255]
.
load_annotations
Davis.load_annotations(sequence, dtype=int)
Arguments
- sequence: String. Sequence name.
- dtype: Numpy Data Type. Data type to return the annotations.
Default value is
np.int
.
Returns
Numpy Array
: Array with the annotations of the given sequence. The
shape of the array will be (nb_frames x H x W)
and the value
will be the index of the objects, being 0
the background.
load_scribble
Davis.load_scribble(sequence, scribble_idx)
Arguments
- sequence: String. Sequence name.
- scribble_idx: Integer. Index of the scribble to load.
Returns
Dictionary
: Scribble data stored in a dictionary with its default
format.