davisinteractive.metrics.jaccard

batched_jaccard

batched_jaccard(y_true,
                y_pred,
                average_over_objects=True,
                nb_objects=None)
Batch jaccard similarity for multiple instance segmentation.

Jaccard similarity over two subsets of binary elements A and B:

\mathcal{J} = \frac{A \cap B}{A \cup B}

Arguments

  • y_true: Numpy Array. Array of shape (B x H x W) and type integer giving the ground truth of the object instance segmentation.
  • y_pred: Numpy Array. Array of shape (B x H x W) and type integer giving the prediction of the object segmentation.
  • average_over_objects: Boolean. Weather or not to average the jaccard over all the objects in the sequence. Default True.
  • nb_objects: Integer. Number of objects in the ground truth mask. If None the value will be infered from y_true. Setting this value will speed up the computation.

Returns

ndarray: Returns an array of shape (B) with the average jaccard for all instances at each frame if average_over_objects=True. If average_over_objects=False returns an array of shape (B x nObj) with nObj being the number of objects on y_true.

batched_f_measure

batched_f_measure(y_true,
                  y_pred,
                  average_over_objects=True,
                  nb_objects=None,
                  bound_th=0.008)
Batch F-measure for multiple instance segmentation.

Arguments

  • y_true: Numpy Array. Array of shape (B x H x W) and type integer giving the ground truth of the object instance segmentation.
  • y_pred: Numpy Array. Array of shape (B x H x W) and type integer giving the prediction of the object segmentation.
  • average_over_objects: Boolean. Weather or not to average the F-measure over all the objects in the sequence. Default True.
  • nb_objects: Integer. Number of objects in the ground truth mask. If None the value will be infered from y_true. Setting this value will speed up the computation.

Returns

ndarray: Returns an array of shape (B) with the average F-measure for all instances at each frame if average_over_objects=True. If average_over_objects=False returns an array of shape (B x nObj) with nObj being the number of objects on y_true.