wormpose.dataset.base_dataset¶
This module contains the WormPose API: abstract classes to subclass in order to add a custom dataset
- class wormpose.dataset.base_dataset.BaseFeaturesDataset[source]¶
Specific code to the dataset to access the features for each video of the dataset
- abstract get_features(video_name: str) dict[source]¶
Returns a dictionary of features
- Returns
dictionary with keys: skeletons, head_width, midbody_width, tail_width, frame_rate, ventral_side, timestamp WHERE skeletons: Coordinates x y of the centerline for each frame in pixel coordinates, a numpy floating point array of shape (N number of frames, J number of joints, 2) The quality of the synthetic images will start degrading when J < 50, consider interpolating if less joints head_width: numpy floating point array of shape N midbody_width: numpy floating point array of shape N tail_width: numpy floating point array of shape N frame_rate: One float number for the frame rate of the video. ventral_side: Optional One string value for the entire video. ‘clockwise’ or ‘anticlockwise’. If None, defaults to anticlockwise timestamp: Optional Timestamp of each frame, a numpy array of shape (N number of frames). If None, will consider each frame to be equidistant in time
- class wormpose.dataset.base_dataset.BaseFramePreprocessing(*args, **kwargs)[source]¶
Specific image processing logic to isolate the worm in the dataset images This object must be pickable (no inner functions for example)
- abstract process(frame: numpy.ndarray) Tuple[numpy.ndarray, int][source]¶
Segment the worm object of interest in the image: returns a mask image of the same shape as frame, where the pixels belong to the worm object of interest are 1, and all the others are 0 Also calculates the average value of the background pixels.
- Parameters
frame – image to process
- Returns
Segmentation mask image , background color
- class wormpose.dataset.base_dataset.BaseFramesDataset[source]¶
Specific code to the dataset to access the frames only. A dataset is divided into several “videos”. Each video contain a list of images or “frames”.
- abstract open(video_name: str)[source]¶
The frames of the dataset are accessed trough a context manager object, in this way we have the option of not entirely loading a big image array in memory if possible
- Parameters
video_name – One video unique id (should be one value of video_names())
- Returns
A context manager object that can be used with the “with” python statement giving access to the frames (array of images) of the dataset Example use : frames = open(“video0”)