dataeval 0.86.0__py3-none-any.whl → 0.86.1__py3-none-any.whl

dataeval/utils/data/metadata.py

@@ -228,58 +228,130 @@ def flatten(
 
     if return_dropped:
         return output, size, _sorted_drop_reasons(dropped)
+    if dropped:
+        dropped_items = "\n".join([f"    {k}: {v}" for k, v in _sorted_drop_reasons(dropped).items()])
+        warnings.warn(f"Metadata entries were dropped:\n{dropped_items}")
+    return output, size
+
+
+def _flatten_for_merge(
+    metadatum: Mapping[str, Any],
+    ignore_lists: bool,
+    fully_qualified: bool,
+    targets: int | None,
+) -> tuple[dict[str, list[Any]] | dict[str, Any], int, dict[str, list[str]]]:
+    flattened, image_repeats, dropped_inner = flatten(
+        metadatum, return_dropped=True, ignore_lists=ignore_lists, fully_qualified=fully_qualified
+    )
+    if targets is not None:
+        # check for mismatch in targets per image and force ignore_lists
+        if not ignore_lists and targets != image_repeats:
+            flattened, image_repeats, dropped_inner = flatten(
+                metadatum, return_dropped=True, ignore_lists=True, fully_qualified=fully_qualified
+            )
+        if targets != image_repeats:
+            flattened = {k: [v] * targets for k, v in flattened.items()}
+        image_repeats = targets
+    return flattened, image_repeats, dropped_inner
+
+
+def _merge(
+    dicts: list[Mapping[str, Any]],
+    ignore_lists: bool,
+    fully_qualified: bool,
+    targets_per_image: Sequence[int] | None,
+) -> tuple[dict[str, list[Any]], dict[str, set[DropReason]], NDArray[np.intp]]:
+    merged: dict[str, list[Any]] = {}
+    isect: set[str] = set()
+    union: set[str] = set()
+    image_repeats = np.zeros(len(dicts), dtype=np.int_)
+    dropped: dict[str, set[DropReason]] = {}
+    for i, d in enumerate(dicts):
+        targets = None if targets_per_image is None else targets_per_image[i]
+        flattened, image_repeats[i], dropped_inner = _flatten_for_merge(d, ignore_lists, fully_qualified, targets)
+        isect = isect.intersection(flattened.keys()) if isect else set(flattened.keys())
+        union.update(flattened.keys())
+        for k, v in dropped_inner.items():
+            dropped.setdefault(k, set()).update({DropReason(vv) for vv in v})
+        for k, v in flattened.items():
+            merged.setdefault(k, []).extend(flattened[k]) if isinstance(v, list) else merged.setdefault(k, []).append(v)
+
+    for k in union - isect:
+        dropped.setdefault(k, set()).add(DropReason.INCONSISTENT_KEY)
+
+    if image_repeats.sum() == image_repeats.size:
+        image_indices = np.arange(image_repeats.size)
     else:
-        if dropped:
-            dropped_items = "\n".join([f"    {k}: {v}" for k, v in _sorted_drop_reasons(dropped).items()])
-            warnings.warn(f"Metadata entries were dropped:\n{dropped_items}")
-        return output, size
+        image_ids = np.arange(image_repeats.size)
+        image_data = np.concatenate(
+            [np.repeat(image_ids[i], image_repeats[i]) for i in range(image_ids.size)], dtype=np.int_
+        )
+        _, image_unsorted = np.unique(image_data, return_inverse=True)
+        image_indices = np.sort(image_unsorted)
 
+    merged = {k: _simplify_type(v) for k, v in merged.items() if k in isect}
+    return merged, dropped, image_indices
 
-def _is_metadata_dict_of_dicts(metadata: Mapping) -> bool:
-    """EXPERIMENTAL: Attempt to detect if metadata is a dict of dicts"""
-    # single dict
-    if len(metadata) < 2:
-        return False
 
-    # dict of non dicts
-    keys = list(metadata)
-    if not isinstance(metadata[keys[0]], Mapping):
-        return False
+@overload
+def merge(
+    metadata: Iterable[Mapping[str, Any]],
+    *,
+    return_dropped: Literal[True],
+    return_numpy: Literal[False] = False,
+    ignore_lists: bool = False,
+    fully_qualified: bool = False,
+    targets_per_image: Sequence[int] | None = None,
+    image_index_key: str = "_image_index",
+) -> tuple[dict[str, list[Any]], dict[str, list[str]]]: ...
+
 
-    # dict of dicts with matching keys
-    return set(metadata[keys[0]]) == set(metadata[keys[1]])
+@overload
+def merge(
+    metadata: Iterable[Mapping[str, Any]],
+    *,
+    return_dropped: Literal[False] = False,
+    return_numpy: Literal[False] = False,
+    ignore_lists: bool = False,
+    fully_qualified: bool = False,
+    targets_per_image: Sequence[int] | None = None,
+    image_index_key: str = "_image_index",
+) -> dict[str, list[Any]]: ...
 
 
 @overload
 def merge(
     metadata: Iterable[Mapping[str, Any]],
+    *,
     return_dropped: Literal[True],
+    return_numpy: Literal[True],
     ignore_lists: bool = False,
     fully_qualified: bool = False,
-    return_numpy: bool = False,
     targets_per_image: Sequence[int] | None = None,
     image_index_key: str = "_image_index",
-) -> tuple[dict[str, list[Any]] | dict[str, NDArray[Any]], dict[str, list[str]]]: ...
+) -> tuple[dict[str, NDArray[Any]], dict[str, list[str]]]: ...
 
 
 @overload
 def merge(
     metadata: Iterable[Mapping[str, Any]],
+    *,
     return_dropped: Literal[False] = False,
+    return_numpy: Literal[True],
     ignore_lists: bool = False,
     fully_qualified: bool = False,
-    return_numpy: bool = False,
     targets_per_image: Sequence[int] | None = None,
     image_index_key: str = "_image_index",
-) -> dict[str, list[Any]] | dict[str, NDArray[Any]]: ...
+) -> dict[str, NDArray[Any]]: ...
 
 
 def merge(
     metadata: Iterable[Mapping[str, Any]],
+    *,
     return_dropped: bool = False,
+    return_numpy: bool = False,
     ignore_lists: bool = False,
     fully_qualified: bool = False,
-    return_numpy: bool = False,
     targets_per_image: Sequence[int] | None = None,
     image_index_key: str = "_image_index",
 ):
@@ -298,12 +370,12 @@ def merge(
         Iterable collection of metadata dictionaries to flatten and merge
     return_dropped: bool, default False
         Option to return a dictionary of dropped keys and the reason(s) for dropping
+    return_numpy : bool, default False
+        Option to return results as lists or NumPy arrays
     ignore_lists : bool, default False
         Option to skip expanding lists within metadata
     fully_qualified : bool, default False
         Option to return dictionary keys full qualified instead of minimized
-    return_numpy : bool, default False
-        Option to return results as lists or NumPy arrays
     targets_per_image : Sequence[int] or None, default None
         Number of targets for each image metadata entry
     image_index_key : str, default "_image_index"
@@ -330,74 +402,24 @@ def merge(
     >>> dropped_keys
     {'target_c': ['inconsistent_key']}
     """
-    merged: dict[str, list[Any]] = {}
-    isect: set[str] = set()
-    union: set[str] = set()
-    keys: list[str] | None = None
-    dicts: list[Mapping[str, Any]]
-
-    # EXPERIMENTAL
-    if isinstance(metadata, Mapping) and _is_metadata_dict_of_dicts(metadata):
-        warnings.warn("Experimental processing for dict of dicts.")
-        keys = [str(k) for k in metadata]
-        dicts = list(metadata.values())
-        ignore_lists = True
-    else:
-        dicts = list(metadata)
+
+    dicts: list[Mapping[str, Any]] = list(metadata)
 
     if targets_per_image is not None and len(dicts) != len(targets_per_image):
         raise ValueError("Number of targets per image must be equal to number of metadata entries.")
 
-    image_repeats = np.zeros(len(dicts), dtype=np.int_)
-    dropped: dict[str, set[DropReason]] = {}
-    for i, d in enumerate(dicts):
-        flattened, image_repeats[i], dropped_inner = flatten(
-            d, return_dropped=True, ignore_lists=ignore_lists, fully_qualified=fully_qualified
-        )
-        if targets_per_image is not None:
-            # check for mismatch in targets per image and force ignore_lists
-            if not ignore_lists and targets_per_image[i] != image_repeats[i]:
-                flattened, image_repeats[i], dropped_inner = flatten(
-                    d, return_dropped=True, ignore_lists=True, fully_qualified=fully_qualified
-                )
-            if targets_per_image[i] != image_repeats[i]:
-                flattened = {k: [v] * targets_per_image[i] for k, v in flattened.items()}
-            image_repeats[i] = targets_per_image[i]
-        isect = isect.intersection(flattened.keys()) if isect else set(flattened.keys())
-        union.update(flattened.keys())
-        for k, v in dropped_inner.items():
-            dropped.setdefault(k, set()).update({DropReason(vv) for vv in v})
-        for k, v in flattened.items():
-            merged.setdefault(k, []).extend(flattened[k]) if isinstance(v, list) else merged.setdefault(k, []).append(v)
-
-    for k in union - isect:
-        dropped.setdefault(k, set()).add(DropReason.INCONSISTENT_KEY)
-
-    if image_repeats.sum() == image_repeats.size:
-        image_indices = np.arange(image_repeats.size)
-    else:
-        image_ids = np.arange(image_repeats.size)
-        image_data = np.concatenate(
-            [np.repeat(image_ids[i], image_repeats[i]) for i in range(image_ids.size)], dtype=np.int_
-        )
-        _, image_unsorted = np.unique(image_data, return_inverse=True)
-        image_indices = np.sort(image_unsorted)
-
-    output: dict[str, Any] = {}
+    merged, dropped, image_indices = _merge(dicts, ignore_lists, fully_qualified, targets_per_image)
 
-    if keys:
-        output["keys"] = np.array(keys) if return_numpy else keys
+    output: dict[str, Any] = {k: np.asarray(v) for k, v in merged.items()} if return_numpy else merged
 
-    for k in (key for key in merged if key in isect):
-        cv = _simplify_type(merged[k])
-        output[k] = np.array(cv) if return_numpy else cv
     if image_index_key not in output:
         output[image_index_key] = image_indices if return_numpy else image_indices.tolist()
 
     if return_dropped:
         return output, _sorted_drop_reasons(dropped)
-    else:
-        if dropped:
-            dropped_items = "\n".join([f"    {k}: {v}" for k, v in _sorted_drop_reasons(dropped).items()])
-            warnings.warn(f"Metadata entries were dropped:\n{dropped_items}")
-        return output
+
+    if dropped:
+        dropped_items = "\n".join([f"    {k}: {v}" for k, v in _sorted_drop_reasons(dropped).items()])
+        warnings.warn(f"Metadata entries were dropped:\n{dropped_items}")
+
+    return output

dataeval/utils/datasets/__init__.py

@@ -1,5 +1,6 @@
 """Provides access to common Computer Vision datasets."""
 
+from dataeval.utils.datasets._antiuav import AntiUAVDetection
 from dataeval.utils.datasets._cifar10 import CIFAR10
 from dataeval.utils.datasets._milco import MILCO
 from dataeval.utils.datasets._mnist import MNIST
@@ -10,6 +11,7 @@ __all__ = [
     "MNIST",
     "Ships",
     "CIFAR10",
+    "AntiUAVDetection",
     "MILCO",
     "VOCDetection",
     "VOCDetectionTorch",

dataeval/utils/datasets/_antiuav.py

@@ -0,0 +1,189 @@
+from __future__ import annotations
+
+__all__ = []
+
+from pathlib import Path
+from typing import TYPE_CHECKING, Any, Literal, Sequence
+
+from defusedxml.ElementTree import parse
+from numpy.typing import NDArray
+
+from dataeval.utils.datasets._base import BaseODDataset, DataLocation
+from dataeval.utils.datasets._mixin import BaseDatasetNumpyMixin
+
+if TYPE_CHECKING:
+    from dataeval.typing import Transform
+
+
+class AntiUAVDetection(BaseODDataset[NDArray[Any]], BaseDatasetNumpyMixin):
+    """
+    A UAV detection dataset focused on detecting UAVs in natural images against large variation in backgrounds.
+
+    The dataset comes from the paper
+    `Vision-based Anti-UAV Detection and Tracking <https://ieeexplore.ieee.org/document/9785379>`_
+    by Jie Zhao et. al. (2022).
+
+    The dataset is approximately 1.3 GB and can be found `here <https://github.com/wangdongdut/DUT-Anti-UAV>`_.
+    Images are collected against a variety of different backgrounds with a variety in the number and type of UAV.
+    Ground truth labels are provided for the train, validation and test set.
+    There are 35 different types of drones along with a variety in lighting conditions and weather conditions.
+
+    There are 10,000 images: 5200 images in the training set, 2200 images in the validation set,
+    and 2600 images in the test set.
+    The dataset only has a single UAV class with the focus being on identifying object location in the image.
+    Ground-truth bounding boxes are provided in (x0, y0, x1, y1) format.
+    The images come in a variety of sizes from 3744 x 5616 to 160 x 240.
+
+    Parameters
+    ----------
+    root : str or pathlib.Path
+        Root directory where the data should be downloaded to or
+        the ``antiuavdetection`` folder of the already downloaded data.
+    image_set: "train", "val", "test", or "base", default "train"
+        If "base", then the full dataset is selected (train, val and test).
+    transforms : Transform, Sequence[Transform] or None, default None
+        Transform(s) to apply to the data.
+    download : bool, default False
+        If True, downloads the dataset from the internet and puts it in root directory.
+        Class checks to see if data is already downloaded to ensure it does not create a duplicate download.
+    verbose : bool, default False
+        If True, outputs print statements.
+
+    Attributes
+    ----------
+    path : pathlib.Path
+        Location of the folder containing the data.
+    image_set : "train", "val", "test", or "base"
+        The selected image set from the dataset.
+    index2label : dict[int, str]
+        Dictionary which translates from class integers to the associated class strings.
+    label2index : dict[str, int]
+        Dictionary which translates from class strings to the associated class integers.
+    metadata : DatasetMetadata
+        Typed dictionary containing dataset metadata, such as `id` which returns the dataset class name.
+    transforms : Sequence[Transform]
+        The transforms to be applied to the data.
+    size : int
+        The size of the dataset.
+
+    Note
+    ----
+    Data License: `Apache 2.0 <https://www.apache.org/licenses/LICENSE-2.0.txt>`_
+    """
+
+    # Need to run the sha256 on the files and then store that
+    _resources = [
+        DataLocation(
+            url="https://drive.usercontent.google.com/download?id=1RVsSGPUKTdmoyoPTBTWwroyulLek1eTj&export=download&authuser=0&confirm=t&uuid=6bca4f94-a242-4bc2-9663-fb03cd94ef2c&at=APcmpox0--NroQ_3bqeTFaJxP7Pw%3A1746552902927",
+            filename="train.zip",
+            md5=False,
+            checksum="14f927290556df60e23cedfa80dffc10dc21e4a3b6843e150cfc49644376eece",
+        ),
+        DataLocation(
+            url="https://drive.usercontent.google.com/download?id=1333uEQfGuqTKslRkkeLSCxylh6AQ0X6n&export=download&authuser=0&confirm=t&uuid=c2ad2f01-aca8-4a85-96bb-b8ef6e40feea&at=APcmpozY-8bhk3nZSFaYbE8rq1Fi%3A1746551543297",
+            filename="val.zip",
+            md5=False,
+            checksum="238be0ceb3e7c5be6711ee3247e49df2750d52f91f54f5366c68bebac112ebf8",
+        ),
+        DataLocation(
+            url="https://drive.usercontent.google.com/download?id=1L1zeW1EMDLlXHClSDcCjl3rs_A6sVai0&export=download&authuser=0&confirm=t&uuid=5a1d7650-d8cd-4461-8354-7daf7292f06c&at=APcmpozLQC1CuP-n5_UX2JnP53Zo%3A1746551676177",
+            filename="test.zip",
+            md5=False,
+            checksum="a671989a01cff98c684aeb084e59b86f4152c50499d86152eb970a9fc7fb1cbe",
+        ),
+    ]
+
+    index2label: dict[int, str] = {
+        0: "unknown",
+        1: "UAV",
+    }
+
+    def __init__(
+        self,
+        root: str | Path,
+        image_set: Literal["train", "val", "test", "base"] = "train",
+        transforms: Transform[NDArray[Any]] | Sequence[Transform[NDArray[Any]]] | None = None,
+        download: bool = False,
+        verbose: bool = False,
+    ) -> None:
+        super().__init__(
+            root,
+            image_set,
+            transforms,
+            download,
+            verbose,
+        )
+
+    def _load_data(self) -> tuple[list[str], list[str], dict[str, list[Any]]]:
+        filepaths: list[str] = []
+        targets: list[str] = []
+        datum_metadata: dict[str, list[Any]] = {}
+
+        # If base, load all resources
+        if self.image_set == "base":
+            metadata_list: list[dict[str, Any]] = []
+
+            for resource in self._resources:
+                self._resource = resource
+                resource_filepaths, resource_targets, resource_metadata = super()._load_data()
+                filepaths.extend(resource_filepaths)
+                targets.extend(resource_targets)
+                metadata_list.append(resource_metadata)
+
+            # Combine metadata
+            for data_dict in metadata_list:
+                for key, val in data_dict.items():
+                    str_key = str(key)  # Ensure key is string
+                    if str_key not in datum_metadata:
+                        datum_metadata[str_key] = []
+                    datum_metadata[str_key].extend(val)
+
+        else:
+            # Grab only the desired data
+            for resource in self._resources:
+                if self.image_set in resource.filename:
+                    self._resource = resource
+                    resource_filepaths, resource_targets, resource_metadata = super()._load_data()
+                    filepaths.extend(resource_filepaths)
+                    targets.extend(resource_targets)
+                    datum_metadata.update(resource_metadata)
+
+        return filepaths, targets, datum_metadata
+
+    def _load_data_inner(self) -> tuple[list[str], list[str], dict[str, Any]]:
+        resource_name = self._resource.filename[:-4]
+        base_dir = self.path / resource_name
+        data_folder = sorted((base_dir / "img").glob("*.jpg"))
+        if not data_folder:
+            raise FileNotFoundError
+
+        file_data = {"image_id": [f"{resource_name}_{entry.name}" for entry in data_folder]}
+        data = [str(entry) for entry in data_folder]
+        annotations = sorted(str(entry) for entry in (base_dir / "xml").glob("*.xml"))
+
+        return data, annotations, file_data
+
+    def _read_annotations(self, annotation: str) -> tuple[list[list[float]], list[int], dict[str, Any]]:
+        """Function for extracting the info for the label and boxes"""
+        boxes: list[list[float]] = []
+        labels = []
+        root = parse(annotation).getroot()
+        if root is None:
+            raise ValueError(f"Unable to parse {annotation}")
+        additional_meta: dict[str, Any] = {
+            "image_width": int(root.findtext("size/width", default="-1")),
+            "image_height": int(root.findtext("size/height", default="-1")),
+            "image_depth": int(root.findtext("size/depth", default="-1")),
+        }
+        for obj in root.findall("object"):
+            labels.append(1 if obj.findtext("name", default="") == "UAV" else 0)
+            boxes.append(
+                [
+                    float(obj.findtext("bndbox/xmin", default="0")),
+                    float(obj.findtext("bndbox/ymin", default="0")),
+                    float(obj.findtext("bndbox/xmax", default="0")),
+                    float(obj.findtext("bndbox/ymax", default="0")),
+                ]
+            )
+
+        return boxes, labels, additional_meta

dataeval/utils/datasets/_base.py

@@ -6,6 +6,8 @@ from abc import abstractmethod
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Generic, Iterator, Literal, NamedTuple, Sequence, TypeVar
 
+import numpy as np
+
 from dataeval.utils.datasets._fileio import _ensure_exists
 from dataeval.utils.datasets._mixin import BaseDatasetMixin
 from dataeval.utils.datasets._types import (
@@ -101,11 +103,7 @@ class BaseDataset(AnnotatedDataset[tuple[_TArray, _TTarget, dict[str, Any]]], Ge
 
     def _get_dataset_dir(self) -> Path:
         # Create a designated folder for this dataset (named after the class)
-        if self._root.stem in [
-            self.__class__.__name__.lower(),
-            self.__class__.__name__.upper(),
-            self.__class__.__name__,
-        ]:
+        if self._root.stem.lower() == self.__class__.__name__.lower():
             dataset_dir: Path = self._root
         else:
             dataset_dir: Path = self._root / self.__class__.__name__.lower()
@@ -114,8 +112,7 @@ class BaseDataset(AnnotatedDataset[tuple[_TArray, _TTarget, dict[str, Any]]], Ge
         return dataset_dir
 
     def _unique_id(self) -> str:
-        unique_id = f"{self.__class__.__name__}_{self.image_set}"
-        return unique_id
+        return f"{self.__class__.__name__}_{self.image_set}"
 
     def _load_data(self) -> tuple[list[str], _TRawTarget, dict[str, Any]]:
         """
@@ -188,6 +185,8 @@ class BaseODDataset(
     Base class for object detection datasets.
     """
 
+    _bboxes_per_size: bool = False
+
     def __getitem__(self, index: int) -> tuple[_TArray, ObjectDetectionTarget[_TArray], dict[str, Any]]:
         """
         Args
@@ -204,8 +203,12 @@ class BaseODDataset(
         boxes, labels, additional_metadata = self._read_annotations(self._targets[index])
         # Get the image
         img = self._read_file(self._filepaths[index])
+        img_size = img.shape
         img = self._transform(img)
-
+        # Adjust labels if necessary
+        if self._bboxes_per_size and boxes:
+            boxes = boxes * np.array([[img_size[1], img_size[2], img_size[1], img_size[2]]])
+        # Create the Object Detection Target
         target = ObjectDetectionTarget(self._as_array(boxes), self._as_array(labels), self._one_hot_encode(labels))
 
         img_metadata = {key: val[index] for key, val in self._datum_metadata.items()}