foxai.explainer.computer_vision.object_detection package

Submodules

foxai.explainer.computer_vision.object_detection.base_object_detector module

File contains abstract base ObjectDetector class.

class foxai.explainer.computer_vision.object_detection.base_object_detector.BaseObjectDetector[source]

Bases: Module, ABC

Base ObjectDetector class which returns predictions with logits to explain.

Code based on https://github.com/pooya-mohammadi/yolov5-gradcam.

abstract forward(image: Tensor) → Tuple[List[PredictionOutput], List[Tensor]][source]

Forward pass of the network.

Parameters:: image – Image to process.
Returns:: Tuple of 2 values, first is tuple of predictions containing bounding-boxes, class number, class name and confidence; second value is list of tensors with logits per each detection.

static preprocessing(img: ndarray, new_shape: Tuple[int, int] = (640, 640), change_original_ratio: bool = False, scaleup: bool = True) → Tensor[source]

Preprocess image before prediction.

Preprocessing is a process consisting of steps: * adding batch dimension * resizing images to desired shapes * adjusting image channels to (B x C x H x W) * convertion to float

Parameters:

img – Image to preprocess.
new_shape – Desired shape of image. Defaults to (640, 640).
change_original_ratio – If resized image should have different height to width ratio than original image. Defaults to False.
scaleup – If scale up image. Defaults to True.

Returns:

Tensor containing preprocessed image.

training: bool

foxai.explainer.computer_vision.object_detection.types module

File contains data class for object detection unified model output.

class foxai.explainer.computer_vision.object_detection.types.ObjectDetectionOutput(saliency_maps: List[Tensor], logits: List[Tensor], predictions: List[PredictionOutput])[source]

Bases: object

Data class for model predictions for object detection.

It contains heatmaps, logits and predictions in YOLO style.

logits: List[Tensor]

predictions: List[PredictionOutput]

saliency_maps: List[Tensor]

class foxai.explainer.computer_vision.object_detection.types.PredictionOutput(bbox: List[int], class_number: int, class_name: str, confidence: float)[source]

Bases: object

Data class for model prediction output in YOLO style.

bbox: List[int]

class_name: str

class_number: int

confidence: float

foxai.explainer.computer_vision.object_detection.utils module

File contains functions to handle images, coordinates and calcualte IoU.

Based on code: https://github.com/pooya-mohammadi/yolov5-gradcam.

foxai.explainer.computer_vision.object_detection.utils.resize_image(image: ndarray, new_shape: Tuple[int, int] = (640, 640), change_original_ratio: bool = False, scaleup: bool = True, interpolation: int = 1) → ndarray[source]

Resize image to given shape.