cv2ext.detection package

Module contents

Submodule for performing simple types of detection.

Classes

AbstractFramePacker

Abstract class for frame packers.

AnnealingFramePacker

A frame packer that uses simulated annealing.

BlobDetector

A simple blob detector class.

RandomFramePacker

A frame packer that randomly samples.

Functions

detect_blobs()

Detect blobs in an image.

draw_detections()

Draw detections on an image.

class cv2ext.detection.AbstractFramePacker[source]

Bases: ABC

Pack regions of a frame together based on detection activity.

Detections are represented as a list of bounding boxes with scores and class id labels optional.

abstract pack(image: np.ndarray, exclude: tuple[int, int, int, int] | list[tuple[int, int, int, int]]) → tuple[np.ndarray, np.ndarray][source]

Pack regions of a frame together.

Parameters:
Returns:

The packed image and the transform information.

Return type:

tuple[np.ndarray, np.ndarray]

abstract unpack(detections: list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]], transform: np.ndarray) → list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]][source]

Unpack regions of a frame.

Parameters:
Return type:

list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]]

abstract update(detections: list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]]) → None[source]

Update the packer with new detections.

Parameters:

detections (list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]]) – The detections to update the packer with.

abstract reset(image_shape: tuple[int, int] | None = None, gridsize: int | None = None) → None[source]

Reset the packer.

Parameters:

  • image_shape (tuple[int, int], optional) – The shape of the image in form (width, height).

  • gridsize (int, optional) – The new gridsize value to use.

class cv2ext.detection.AbstractGridFramePacker(image_shape: tuple[int, int], gridsize: int = 128, detection_buffer: int = 30, method: str = 'shelf')[source]

Bases: AbstractFramePacker

Pack regions of a frame together based on a grid.

reset(image_shape: tuple[int, int] | None = None, gridsize: int | None = None) → None[source]

Reset the packer.

Parameters:

  • image_shape (tuple[int, int], optional) – The shape of the image in form (width, height).

  • gridsize (int, optional) – The new gridsize value to use.

pack(image: ndarray, exclude: tuple[int, int, int, int] | list[tuple[int, int, int, int]] | None = None, method: str | None = None) → tuple[ndarray, ndarray][source]

Pack regions of a frame together.

Parameters:

  • image (np.ndarray) – The image to pack.

  • exclude (tuple[int, int, int, int] | list[tuple[int, int, int, int]], optional) – Regions of the image to exclude from the packing. By default None.

  • method (str, optional) – The method to pack the bounding boxes with. By default, None Options are: [‘simple’, ‘shelf’]

Returns:

The packed image and the transform information.

Return type:

tuple[np.ndarray, np.ndarray]

Raises:

ValueError – If the image shape does not match the packer shape.

unpack(detections: list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]], transform: np.ndarray) → list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]][source]

Unpack regions of a frame.

Parameters:
Return type:

list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]]

update(detections: list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]]) → None[source]

Update the packer with new detections.

Parameters:

detections (list[tuple[int, int, int, int]] | list[tuple[tuple[int, int, int, int], float, int]]) – The detections to update the packer with.

class cv2ext.detection.AnnealingFramePacker(image_shape: tuple[int, int], gridsize: int = 128, alpha: float = 0.01, min_prob: float = 0.1, detection_buffer: int = 30, method: str = 'shelf')[source]

Bases: AbstractGridFramePacker

Pack regions of a frame together based on detection activity.

Detections are represented as a list of bounding boxes with scores and class id labels optional.

class cv2ext.detection.BlobDetector(kernel_size: int = 9, sigma: float = 2.0, max_area: float = 0.9, min_area: float = 0.05, *, use_blur: bool | None = None, is_rgb: bool | None = None, filter_size: bool | None = None)[source]

Bases: object

Class for detecting blobs in images.

detect(image: np.ndarray) → list[tuple[int, int, int, int]][source]

Detect blobs in an image.

Parameters:

image (np.ndarray) – The image to detect blobs in.

Returns:

A list of bounding boxes for each blob.

Return type:

list[tuple[int, int, int, int]]

class cv2ext.detection.RandomFramePacker(image_shape: tuple[int, int], gridsize: int = 128, threshold: float = 0.1, detection_buffer: int = 30, method: str = 'shelf')[source]

Bases: AbstractGridFramePacker

Pack regions of a frame together randomly.

cv2ext.detection.detect_blobs(image: np.ndarray, kernel_size: int = 9, sigma: float = 2.0, max_area: float = 0.9, min_area: float = 0.05, *, use_blur: bool | None = None, is_rgb: bool | None = None, filter_size: bool | None = None) → list[tuple[int, int, int, int]][source]

Detect blobs in an image.

Blob detection is performed using OpenCVs, findContours function. Each contour is identified and then a bounding box is created for each one.

Parameters:

  • image (np.ndarray) – The image to detect blobs in.

  • kernel_size (int) – The size of the kernel for the Gaussian blur.

  • sigma (float) – The standard deviation for the Gaussian blur.

  • max_area (float) – The maximum area a bounding box can be relative to the image. Default is 0.9, so bounding boxes larger than 90% of the image will be removed from final result if filtering enabled.

  • min_area (float) – The minimum area a bounding box can be relative to the image. Default is 0.01, so bounding boxes smaller than 5% of the image will be removed from final result if filtering enabled.

  • use_blur (bool, optional) – Whether or not to use a Gaussian blur on the image. By default, blur will be used.

  • is_rgb (bool, optional) – Whether or not the image is in RGB format. Pass True if input image is RGB instead of BGR (OpenCV default).

  • filter_size (bool, optional) – Whether or not to filter the final bounding boxes by the area of the bounding box.

Returns:

A list of bounding boxes for each blob.

Return type:

list[tuple[int, int, int, int]]

cv2ext.detection.draw_detections(image: np.ndarray, dets: Sequence[tuple[tuple[int, int, int, int], float, int]], class_map: dict[int, str] | None = None, color: Color | tuple[int, int, int] = Color.RED, thickness: int = 2, opacity: float | None = None, *, copy: bool | None = None) → np.ndarray[source]

Draw bounding boxes on an image.

Parameters:

  • image (np.ndarray) – The image to draw the bounding boxes on.

  • dets (Sequence[tuple[tuple[int, int, int, int], float, int]]) – The detections to draw. The detections should be in form: (bbox, confidence, classid)

  • class_map (dict[int, str], optional) – The class map to use for converting class indices to labels.

  • color (Color, tuple[int, int, int], optional) – The color to draw the bounding boxes. In BGR format and the default is Color.RED.

  • thickness (int, optional) – The thickness of the bounding box lines. Default is 2.

  • opacity (float, optional) – The opacity to draw the bounding boxes with. By default None or 100%. Can have a high performance impact.

  • copy (bool, optional) – Whether or not to copy the image before drawing. Default is False.

Returns:

The image with the bounding boxes drawn.

Return type:

np.ndarray