edgefirst-validator 4.2.1__py3-none-any.whl

edgefirst/validator/datasets/instance/__init__.py

@@ -0,0 +1,4 @@
+from edgefirst.validator.datasets.instance.core import Instance
+from edgefirst.validator.datasets.instance.detection import DetectionInstance
+from edgefirst.validator.datasets.instance.segmentation import SegmentationInstance
+from edgefirst.validator.datasets.instance.multitask import MultitaskInstance

edgefirst/validator/datasets/instance/core.py

@@ -0,0 +1,222 @@
+import numpy as np
+
+
+class Instance:
+    """
+    Base Instance object for containing the ground truth dataset instance.
+
+    Parameters
+    ----------
+    image_path: str
+        The path to the image for Darknet datasets. Otherwise this is the
+        image name for TFRecord datasets. This is required either to
+        allow reading the image from file or saving the image results
+        in disk with the same file name.
+    """
+
+    def __init__(self, image_path: str):
+        self.__image_path = image_path
+        # This is the NumPy array image.
+        self.__image = None
+        self.__visual_image = None
+        self.__height = 0
+        self.__width = 0
+
+        # Same property used in YOLOv5 and YOLOv7 for the shapes.
+        self.__shapes = [
+            [
+                [0, 0],  # imgsz (model input shape) [height, width]
+                # ratio_pad [[scale y, scale x], [pad w, pad h]]
+                [(1.0, 1.0), (0.0, 0.0)]
+            ],
+            [1.0, 1.0]  # label ratio [x, y]
+        ]
+        # The original image dimensions.
+        self.__image_shape = None
+
+    @property
+    def image_path(self) -> str:
+        """
+        Attribute to access the image path/name.
+
+        Returns
+        -------
+        str
+            The image path/name.
+        """
+        return self.__image_path
+
+    @image_path.setter
+    def image_path(self, path: str):
+        """
+        Sets the image path/name.
+
+        Parameters
+        ----------
+        path: str
+            The image path/name.
+        """
+        self.__image_path = path
+
+    @property
+    def image(self) -> np.ndarray:
+        """
+        Attribute to access the image.
+
+        Returns
+        -------
+        np.ndarray
+            The image as a NumPy array.
+        """
+        return self.__image
+
+    @image.setter
+    def image(self, this_image: np.ndarray):
+        """
+        Sets the image array.
+
+        Parameters
+        ----------
+        this_image: np.ndarray
+            The image array.
+        """
+        self.__image = this_image
+
+    @property
+    def visual_image(self) -> np.ndarray:
+        """
+        Attribute to access the image.
+
+        Returns
+        -------
+        np.ndarray
+            The image as a NumPy array.
+        """
+        return self.__visual_image
+
+    @visual_image.setter
+    def visual_image(self, this_image: np.ndarray):
+        """
+        Sets the image array.
+
+        Parameters
+        ----------
+        this_image: np.ndarray
+            The image array.
+        """
+        self.__visual_image = this_image
+
+    @property
+    def height(self) -> int:
+        """
+        Attribute to access the image height in pixels.
+
+        Returns
+        -------
+        int
+            The image height in pixels. 0 means uninitialized.
+        """
+        return self.__height
+
+    @height.setter
+    def height(self, image_height: int):
+        """
+        Sets the image height dimension to a new value.
+
+        Parameters
+        ----------
+        image_height: int
+            This is the new image height to set.
+        """
+        self.__height = image_height
+
+    @property
+    def width(self) -> int:
+        """
+        Attribute to access the image width in pixels.
+
+        Returns
+        -------
+        int:
+            The image width in pixels. 0 means uninitialized.
+        """
+        return self.__width
+
+    @width.setter
+    def width(self, image_width: int):
+        """
+        Sets the image width dimension to a new value.
+
+        Parameters
+        ----------
+        image_width: int
+            This is the new image width to set.
+        """
+        self.__width = image_width
+
+    @property
+    def shapes(self) -> list:
+        """
+        Attribute to access the inference shapes which includes
+        the image shape after letterbox operation and aswell as padding
+        and ratio which are variables that mirrors YOLOv5 and YOLOv7
+        implementation of the letterbox image processing.
+        This attribute is used for bounding box rescaling for both the
+        ground truth and detections.
+
+        Returns
+        -------
+        list
+            .. code-block:: python
+
+                [
+                    [image height, image width]],
+                    [[ratio y, ratio x], [pad x, pad y]],
+                    label_ratio
+                ]
+        """
+        return self.__shapes
+
+    @shapes.setter
+    def shapes(self, shape: list):
+        """
+        Sets the shapes attribute to a new value.
+
+        Parameters
+        ----------
+        shape: list
+            .. code-block:: python
+
+                [
+                    [image height, image width]],
+                    [[ratio y, ratio x], [pad x, pad y]],
+                    label_ratio
+                ]
+        """
+        self.__shapes = shape
+
+    @property
+    def image_shape(self) -> tuple:
+        """
+        Attribute to access the image shape.
+        This is the original image dimensions prior
+        to image preprocessing.
+
+        Returns
+        -------
+        tuple
+            This contains the image (height, width) dimensions.
+        """
+        return self.__image_shape
+
+    @image_shape.setter
+    def image_shape(self, shape: tuple):
+        """
+        Sets the image shape.
+
+        Parameters
+        ----------
+        shape: tuple
+            Sets the original image dimensions.
+        """
+        self.__image_shape = shape

edgefirst/validator/datasets/instance/detection.py

@@ -0,0 +1,145 @@
+from typing import Union
+
+import numpy as np
+
+from edgefirst.validator.datasets.instance import Instance
+
+
+class DetectionInstance(Instance):
+    """
+    Instance for storing ground truth and
+    model bounding boxes, labels, and scores.
+
+    Parameters
+    ----------
+    image_path: str
+        The path to the image for Darknet datasets. Otherwise this is the
+        image name for TFRecord datasets. This is required either to
+        allow reading the image from file or saving the image results
+        in disk with the same file name.
+    """
+
+    def __init__(self, image_path: str):
+        super(DetectionInstance, self).__init__(image_path)
+
+        # These are the 2D bounding boxes in either YOLO, COCO, or PascalVoc.
+        self.__boxes = np.array([])
+        # These contain either string or integer labels per bounding box.
+        self.__labels = np.array([])
+        # These contain the prediction scores per bounding box. Empty if gt.
+        self.__scores = np.array([])
+
+    @property
+    def boxes(self) -> Union[list, np.ndarray]:
+        """
+        Attribute to access the 2D bounding boxes for detection.
+
+        Returns
+        -------
+        Union[list, np.ndarray]
+            This contains the 2D normalized bounding boxes.
+        """
+        return self.__boxes
+
+    @boxes.setter
+    def boxes(self, boxes_2d: Union[list, np.ndarray]):
+        """
+        Sets the 2D bounding boxes to a new value.
+
+        Parameters
+        ----------
+        boxes_2d: Union[list, np.ndarray]
+            These are the 2D bounding boxes to set.
+        """
+        self.__boxes = boxes_2d
+
+    def append_boxes(self, box: Union[list, np.ndarray]):
+        """
+        Appends list or stacks NumPy array 2D bounding boxes.
+
+        Parameters
+        ----------
+        box: Union[list, np.ndarray]
+            This is the 2D normalized bounding box in either
+            YOLO, COCO, or PascalVoc.
+        """
+        if isinstance(self.__boxes, np.ndarray):
+            self.__boxes = np.vstack([self.__boxes, box])
+        else:
+            self.__boxes.append(box)
+
+    @property
+    def labels(self) -> Union[list, np.ndarray]:
+        """
+        Attribute to access the labels per bounding box.
+
+        Returns
+        -------
+        Union[list, np.ndarray]
+            This contains the labels per bounding box.
+        """
+        return self.__labels
+
+    @labels.setter
+    def labels(self, new_labels: Union[list, np.ndarray]):
+        """
+        Sets the labels to a new value.
+
+        Parameters
+        ----------
+        new_labels: Union[list, np.ndarray]
+            These are the labels to set.
+        """
+        self.__labels = new_labels
+
+    def append_labels(self, label: Union[str, int, np.integer]):
+        """
+        Appends list or appends NumPy array label.
+
+        Parameters
+        ----------
+        label: Union[str, int, np.integer]
+            This is the label to append to the list.
+        """
+        if isinstance(self.__labels, np.ndarray):
+            self.__labels = np.append(self.__labels, label)
+        else:
+            self.__labels.append(label)
+
+    @property
+    def scores(self) -> Union[list, np.ndarray]:
+        """
+        Attribute to access the scores per bounding box.
+
+        Returns
+        -------
+        Union[list, np.ndarray]
+            This contains the scores per bounding box.
+        """
+        return self.__scores
+
+    @scores.setter
+    def scores(self, new_scores: Union[list, np.ndarray]):
+        """
+        Sets the scores to a new value.
+
+        Parameters
+        ----------
+        new_scores: Union[list, np.ndarray]
+            These are the scores to set.
+        """
+        self.__scores = new_scores
+
+    def append_scores(self, score: float):
+        """
+        Appends list or appends NumPy array scores.
+
+        Parameters
+        ----------
+        score: float
+            This is the score to append to the list.
+        """
+        if isinstance(self.__scores, np.ndarray):
+            self.__scores = np.append(self.__scores, score)
+        else:
+            self.__scores.append(score)

edgefirst/validator/datasets/instance/multitask.py

@@ -0,0 +1,80 @@
+from typing import Union
+
+import numpy as np
+
+from edgefirst.validator.datasets.instance import DetectionInstance
+
+
+class MultitaskInstance(DetectionInstance):
+    """
+    Instance for storing Multitask ground truth and model predictions
+    for Vision which has bounding boxes, labels, scores, and masks.
+
+    Parameters
+    ----------
+    image_path: str
+        The path to the image for Darknet datasets. Otherwise this is the
+        image name for TFRecord datasets. This is required either to
+        allow reading the image from file or saving the image results
+        in disk with the same file name.
+    """
+
+    def __init__(self, image_path: str):
+        super(MultitaskInstance, self).__init__(image_path)
+
+        # These contain the segmentation points to form the
+        # polygon shape around the object.
+        self.__polygons = np.array([])
+        # This is the segmentation mask for the image.
+        self.__mask = None
+
+    @property
+    def polygons(self) -> Union[list, np.ndarray]:
+        """
+        Attribute to access the 2D points that form the polygon to shape
+        around the object to form segmentation masks.
+
+        Returns
+        -------
+        Union[list, np.ndarray]
+            This contains the polygon points. Ex.
+            [[[x1,y1], [x2,y2], ... ,[xn,yn]], [...], ...]
+        """
+        return self.__polygons
+
+    @polygons.setter
+    def polygons(self, new_polygons: Union[list, np.ndarray]):
+        """
+        Sets the polygons to a new value.
+
+        Parameters
+        ----------
+        new_polygons: Union[list, np.ndarray]
+            This is the polygons to set.
+        """
+        self.__polygons = new_polygons
+
+    @property
+    def mask(self) -> np.ndarray:
+        """
+        Attribute to access the segmentation mask of the image.
+
+        Returns
+        -------
+        np.ndarray
+            This contains the mask with the same dimensions as the image
+            providing an integer label per pixel to represent the mask.
+        """
+        return self.__mask
+
+    @mask.setter
+    def mask(self, new_mask: np.ndarray):
+        """
+        Sets the mask to a new value.
+
+        Parameters
+        ----------
+        new_mask: np.ndarray
+            This is the mask to set.
+        """
+        self.__mask = new_mask

edgefirst/validator/datasets/instance/segmentation.py

@@ -0,0 +1,120 @@
+from typing import Union
+
+import numpy as np
+
+from edgefirst.validator.datasets.instance import Instance
+
+
+class SegmentationInstance(Instance):
+    """
+    Instance for storing segmentation properties from the
+    ground truth or model predictions.
+
+    Parameters
+    ----------
+    image_path: str
+        The path to the image for Darknet datasets. Otherwise this is the
+        image name for TFRecord datasets. This is required either to
+        allow reading the image from file or saving the image results
+        in disk with the same file name.
+    """
+
+    def __init__(self, image_path: str):
+        super(SegmentationInstance, self).__init__(image_path)
+
+        # These are the unique integer labels in the segmentation mask.
+        self.__labels = np.array([])
+        # These contain the segmentation points to form the
+        # polygon shape around the object.
+        self.__polygons = np.array([])
+        # This is the segmentation mask for the image.
+        self.__mask = None
+
+    @property
+    def labels(self) -> Union[list, np.ndarray]:
+        """
+        Attribute to access the labels per bounding box.
+
+        Returns
+        -------
+        Union[list, np.ndarray]
+            This contains the labels per bounding box.
+        """
+        return self.__labels
+
+    @labels.setter
+    def labels(self, new_labels: Union[list, np.ndarray]):
+        """
+        Sets the labels to a new value.
+
+        Parameters
+        ----------
+        new_labels: Union[list, np.ndarray]
+            These are the labels to set.
+        """
+        self.__labels = new_labels
+
+    def append_labels(self, label: Union[str, int, np.integer]):
+        """
+        Appends list or appends NumPy array label.
+
+        Parameters
+        ----------
+        label: Union[str, int, np.integer]
+            This is the label to append to the list.
+        """
+        if isinstance(self.__labels, np.ndarray):
+            self.__labels = np.append(self.__labels, label)
+        else:
+            self.__labels.append(label)
+
+    @property
+    def polygons(self) -> Union[list, np.ndarray]:
+        """
+        Attribute to access the 2D points that form the polygon to shape
+        around the object to form segmentation masks.
+
+        Returns
+        -------
+        Union[list, np.ndarray]
+            This contains the polygon points. Ex.
+            [[[x1,y1], [x2,y2], ... ,[xn,yn]], [...], ...]
+        """
+        return self.__polygons
+
+    @polygons.setter
+    def polygons(self, new_polygons: Union[list, np.ndarray]):
+        """
+        Sets the polygons to a new value.
+
+        Parameters
+        ----------
+        new_polygons: Union[list, np.ndarray]
+            This is the polygons to set.
+        """
+        self.__polygons = new_polygons
+
+    @property
+    def mask(self) -> np.ndarray:
+        """
+        Attribute to access the segmentation mask of the image.
+
+        Returns
+        -------
+        np.ndarray
+            This contains the mask with the same dimensions as the image
+            providing an integer label per pixel to represent the mask.
+        """
+        return self.__mask
+
+    @mask.setter
+    def mask(self, new_mask: np.ndarray):
+        """
+        Sets the mask to a new value.
+
+        Parameters
+        ----------
+        new_mask: np.ndarray
+            This is the mask to set.
+        """
+        self.__mask = new_mask