cap_bounding_box_coordinates

Dataset.cap_bounding_box_coordinates() → Self

Method to ensure the bounding box coordinates are inside the picture frame. Indeed, some dataset (like crowdhuman) do use outside of picture bounding box

Returns:

New Dataset with bounding box capped so that X and Y coordinates are inside corresponding picture dimensions

Example

>>> from lours.utils.doc_utils import dummy_dataset
>>> from lours.utils.testing import assert_bounding_boxes_well_formed
>>> example = dummy_dataset(1, 1)
>>> example.annotations.loc[0, "box_y_min"] = -0.5
>>> example.annotations.loc[0, "box_height"] = (
...     example.images["height"][0] + 1
... )
>>> example
Dataset object containing 1 image and 1 object
Name :
    inside_else_memory
Images root :
    such/serious
Images :
    width  height relative_path   type  split
id
0     342     377  help/me.jpeg  .jpeg  train
Annotations :
    image_id category_str  category_id  ... box_y_min   box_width  box_height
id                                      ...
0          0    interview           25  ...      -0.5  306.509956       378.0

[1 rows x 8 columns]
Label map :
{15: 'step', 19: 'why', 25: 'interview'}

>>> assert_bounding_boxes_well_formed(example)
Traceback (most recent call last):
    ...
AssertionError: Assertion failed. Bounding boxes must have positive Y values. First occurrence at row 0 : image_id                0
category_str     interview
category_id             25
split                train
box_x_min         5.652451
box_y_min             -0.5
box_width       306.509956
box_height           378.0
Name: 0, dtype: object

>>> example.cap_bounding_box_coordinates()
Dataset object containing 1 image and 1 object
Name :
    inside_else_memory
Images root :
    such/serious
Images :
    width  height relative_path   type  split
id
0     342     377  help/me.jpeg  .jpeg  train
Annotations :
    image_id category_str  category_id  ... box_y_min   box_width  box_height
id                                      ...
0          0    interview           25  ...       0.0  306.509956       377.0

[1 rows x 8 columns]
Label map :
{15: 'step', 19: 'why', 25: 'interview'}