to_custom_displayable_masks

hybrid_learning.experimentation.fuzzy_exp.fuzzy_exp_vis.to_custom_displayable_masks(out, person_key='pedestrian', skip_if_max_lower_than=None, mark_if_score_higher_than=None, scores_to_add_to_key=None, reduce_masks=None, compare_masks=None, mask_union=None)[source]

Custom post-process of a formula calculation output dict for easy display with compare_orig_with_masks. Modifications applied:

  • stacked masks are unstacked or reduced via mask union

  • the person mask(s) potentially get scores added to their keys according to scores_to_add_to_key

  • the person mask(s) potentially get marked with color if condition mark_if_score_higher_than is met

  • in case conditions are not met (skip_if_max_lower_than), None is returned

Assumptions:

  • tensor of len(shape)==1 is a list of scores (one per prediction)

  • tensor of len(shape)==2 is a standard masks

  • tensor of len(shape)==3 are standard masks stacked in dim 0

Parameters

  • out (Dict[str, Tensor]) – the dict output of a formula evaluation; only tensors values therein are used

  • person_key (str) – the key of the (stacked) person mask(s)

  • skip_if_max_lower_than (Optional[Dict[str, float]]) – dict of the form {tensor_key: minimum_value_of_max}; return None if the max value of tensors in out at keys are lower than the minimum_value_of_max

  • scores_to_add_to_key (Optional[Sequence[Sequence[str]]]) – add values of given scores to the string key(s) of the (unstacked) person mask(s)

  • reduce_masks (Optional[Sequence[str]]) – reduce the masks at given keys if they are stacked; defaults to all stacked masks except for the person mask(s)

  • compare_masks (Optional[Sequence[Sequence[str]]]) – each item is a list of keys; for each item add a comparison image to the output comparing the masks at keys in that item; if the key of a later unstacked mask is given, the unstacked masks are merged via union for comparison

  • mark_if_score_higher_than (Optional[Dict[str, float]]) –

  • mask_union (Optional[Callable[[Sequence[Tensor]], Tensor]]) –

Returns

dict of {title: mask} of 2D and 3D tensors representing masks and images for plotting

Return type

Dict[str, Tensor]