inet.models package¶
Subpackages¶
Submodules¶
inet.models.data_structures module¶
- class BoundingBox(x: float, y: float, w: float, h: float)[source]¶
Bases:
object
BoundingBox representation to parse, draw and evaluate
- A_I(other: inet.models.data_structures.BoundingBox) float [source]¶
Area of intersection with another BBox other :param other: a different BBox :return: the area of intersection
- A_U(other: inet.models.data_structures.BoundingBox) float [source]¶
Area of union with other BBox other :param other: a different BBox :return: the area of union
- GIoU(other: inet.models.data_structures.BoundingBox) float [source]¶
Generalized intersection over union (GIoU) based on [the paper](https://giou.stanford.edu/GIoU.pdf) :param other: other BB to compute GIoU with :return: GIoU for self and other
- IoU(other: inet.models.data_structures.BoundingBox) float [source]¶
Computes the intersection over union (IoU) with a different BBox other :param other: a different BBox :return: intersection over union value
- property area: float¶
- Returns
- draw(gc, color: str = 'red') None [source]¶
Method to render BBox into a graphic-context gc :param gc: graphics-context, e.g. matplotlib.pyplot.gca() :param color: the color of the BBox :return:
- property half_h: float¶
- Returns
half of the bbs height
- property half_w: float¶
- Returns
half of the bbs width
- overlap(bb2: inet.models.data_structures.BoundingBox) Optional[inet.models.data_structures.BoundingBox] [source]¶
Generates an overlapping BBox/convex hull around self and bb2 :param bb2: a different BBox :return: when overlapping, a new BBox containing both BBoxes
- property x_max: float¶
- Returns
x-max coordinate
- property y_max: float¶
- Returns
y-max coordinate
- class ModelArchitecture(backbone: tensorflow.python.keras.engine.sequential.Sequential, name: str, create_model: Optional[Callable[[tensorflow.python.keras.engine.sequential.Sequential, Optional[str], Optional[int], Optional[float]], inet.models.architectures.base_model.TaskModel]] = None)[source]¶
Bases:
object
Helper dataclass to simplify creation of model architecture, mostly used in development notebooks.
- backbone: tensorflow.python.keras.engine.sequential.Sequential¶
- create_model: Optional[Callable[[tensorflow.python.keras.engine.sequential.Sequential, Optional[str], Optional[int], Optional[float]], inet.models.architectures.base_model.TaskModel]] = None¶
- name: str¶
inet.models.hyper_parameter_optimization module¶
- class FrozenBlockConf(value)[source]¶
Bases:
enum.Enum
Helper to freeze layers in a sequential model
- TRAIN_ALL = 0¶
- TRAIN_HALF = 1¶
- TRAIN_NONE = 2¶
- read_trials(dir_name: str) List [source]¶
Reads trial files provided by [keras-tuner](https://keras.io/keras_tuner/).