Stanford cars

docs/source/datasets.rst

@@ -70,6 +70,7 @@ You can also create your own datasets using the provided :ref:`base classes <bas
     SBU
     SEMEION
     Sintel
+    StanfordCars
     STL10
     SUN397
     SVHN

test/test_datasets.py

@@ -2463,5 +2463,61 @@ def _meta_to_split_and_classification_ann(self, meta, idx):
         return (image_id, class_id, species, breed_id)
 
 
+class StanfordCarsTestCase(datasets_utils.ImageDatasetTestCase):
+    DATASET_CLASS = datasets.StanfordCars
+    REQUIRED_PACKAGES = ("scipy",)
+    ADDITIONAL_CONFIGS = datasets_utils.combinations_grid(train=(True, False))
+    FEATURE_TYPES = (PIL.Image.Image, int)
+
+    def _inject_fake_data(self, tmpdir, config):
+        import scipy.io as io
+        from numpy.core.records import fromarrays  # to record arrays similar to matlab format
+
+        train = config["train"]
+        num_examples = 5
+        root_folder = os.path.join(tmpdir, "stanforddcars")
+        os.makedirs(root_folder, exist_ok=True)
+
+        # generate random data for labels
+        class_name = np.random.randint(0, 100, num_examples, dtype=np.uint8)
+        bbox_x1 = np.random.randint(0, 100, num_examples, dtype=np.uint8)
+        bbox_x2 = np.random.randint(0, 100, num_examples, dtype=np.uint8)
+
+        bbox_y1 = np.random.randint(0, 100, num_examples, dtype=np.uint8)
+        bb1ox_y2 = np.random.randint(0, 100, num_examples, dtype=np.uint8)
+        fname = [f"{i:5d}.jpg" for i in range(num_examples)]
+
+        rec_array = fromarrays(
+            [bbox_x1, bbox_y1, bbox_x2, bb1ox_y2, class_name, fname],
+            names=["bbox_x1", "bbox_y1", "bbox_x2", "bbox_y2", "class", "fname"],
+        )
+
+        if train:
+            # create training image folder
+            datasets_utils.create_image_folder(
+                root=root_folder,
+                name="cars_train",
+                file_name_fn=lambda image_index: f"{image_index:5d}.jpg",
+                num_examples=num_examples,
+            )
+            devkit = os.path.join(root_folder, "devkit")
+            io.savemat(
+                f"{devkit}/cars_train_annos.mat", {"annotations": rec_array}
+            )  # save the recorded array as matlab file
+        else:
+            # create test image folder
+            datasets_utils.create_image_folder(
+                root=root_folder,
+                name="cars_test",
+                file_name_fn=lambda image_index: f"{image_index:5d}.jpg",
+                num_examples=num_examples,
+            )
+            io.savemat(
+                f"{root_folder}/cars_test_annos_withlabels.mat", {"annotations": rec_array}
+            )  # save recorded array as matlab file
+
+        return num_examples
+
+
 if __name__ == "__main__":
     unittest.main()

torchvision/datasets/__init__.py

@@ -27,6 +27,7 @@
 from .sbd import SBDataset
 from .sbu import SBU
 from .semeion import SEMEION
+from .standford_cars import StanfordCars
 from .stl10 import STL10
 from .sun397 import SUN397
 from .svhn import SVHN
@@ -51,6 +52,7 @@
     "QMNIST",
     "MNIST",
     "KMNIST",
+    "StanfordCars",
     "STL10",
     "SUN397",
     "SVHN",

torchvision/datasets/standford_cars.py

@@ -0,0 +1,145 @@
+import os
+import os.path
+from typing import Callable, Optional
+
+from PIL import Image
+
+from .utils import download_and_extract_archive, download_url
+from .vision import VisionDataset
+
+
+class StanfordCars(VisionDataset):
+    """`Stanford Cars <https://ai.stanford.edu/~jkrause/cars/car_dataset.html>`_ Dataset
+
+    .. warning::
+
+        This class needs `scipy <https://docs.scipy.org/doc/>`_ to load target files from `.mat` format.
+
+    Args:
+        root (string): Root directory of dataset
+        train (bool, optional):If True, creates dataset from training set, otherwise creates from test set
+        transform (callable, optional): A function/transform that  takes in an PIL image
+            and returns a transformed version. E.g, ``transforms.RandomCrop``
+        target_transform (callable, optional): A function/transform that takes in the
+            target and transforms it.
+        download (bool, optional): If True, downloads the dataset from the internet and
+            puts it in root directory. If dataset is already downloaded, it is not
+            downloaded again."""
+
+    urls = (
+        "https://ai.stanford.edu/~jkrause/car196/cars_test.tgz",
+        "https://ai.stanford.edu/~jkrause/car196/cars_train.tgz",
+    )  # test and train image urls
+
+    md5s = (
+        "4ce7ebf6a94d07f1952d94dd34c4d501",
+        "065e5b463ae28d29e77c1b4b166cfe61",
+    )  # md5checksum for test and train data
+
+    annot_urls = (
+        "https://ai.stanford.edu/~jkrause/car196/cars_test_annos_withlabels.mat",
+        "https://ai.stanford.edu/~jkrause/cars/car_devkit.tgz",
+    )  # annotations and labels for test and train
+
+    annot_md5s = (
+        "b0a2b23655a3edd16d84508592a98d10",
+        "c3b158d763b6e2245038c8ad08e45376",
+    )  # md5 checksum for annotations
+
+    def __init__(
+        self,
+        root: str,
+        train: bool = True,
+        transform: Optional[Callable] = None,
+        target_transform: Optional[Callable] = None,
+        download: bool = False,
+    ) -> None:
+
+        try:
+            from scipy.io import loadmat
+
+            self._loadmat = loadmat
+        except ImportError:
+            raise RuntimeError("Scipy is not found. This dataset needs to have scipy installed: pip install scipy")
+
+        super().__init__(root, transform=transform, target_transform=target_transform)
+
+        self.train = train
+
+        if download:
+            self.download()
+
+        if not self._check_exists():
+            raise RuntimeError("Dataset not found. You can use download=True to download it")
+
+        self._samples = self._make_dataset()
+        self.classes = self._get_classes_name()  # class_id to class_name mapping
+
+    def _get_class_names(self) -> dict:
+        """
+        Returns Mapping of class ids to class names in form of Dictionary
+        """
+        meta_data = self._loadmat(os.path.join(self.root, "devkit/cars_meta.mat"))
+        class_names = meta_data["class_names"][0]
+        return {class_name[0].replace(" ", "_").replace("/", "_"): i for i, class_name in enumerate(class_names)}
+
+    def _make_dataset(self):
+        """
+        Returns Annotations for training data and testing data
+        """
+        annotations = None
+        if self.train:
+            annotations = self._loadmat(os.path.join(self.root, "devkit/cars_train_annos.mat"))
+        else:
+            annotations = self._loadmat(os.path.join(self.root, "cars_test_annos_withlabels.mat"))
+        samples = []
+        annotations = annotations["annotations"][0]
+        for index in range(len(annotations)):
+            target = annotations[index][4][0, 0]
+            image_file = annotations[index][5][0]
+            # Beware: Stanford cars targets starts at 1
+            target = target - 1
+            samples.append((image_file, target))
+        return samples
+
+    def __len__(self) -> int:
+        return len(self._samples)
+
+    def __getitem__(self, idx: int) -> (Image, int):
+        """Returns pil_image and class_id for given index"""
+        image_file, target = self._samples[idx]
+        image_path = os.path.join(self.root, f"cars_{'train' if self.train else 'test'}", image_file)
+        pil_image = Image.open(image_path).convert("RGB")
+
+        if self.transform is not None:
+            pil_image = self.transform(pil_image)
+        if self.target_transform is not None:
+            target = self.target_transform(target)
+        return pil_image, target
+
+    def download(self) -> None:
+        if self._check_exists():
+            return
+        else:
+            download_and_extract_archive(
+                url=self.urls[self.train], download_root=self.root, extract_root=self.root, md5=self.md5s[self.train]
+            )
+            download_and_extract_archive(
+                url=self.annot_urls[1], download_root=self.root, extract_root=self.root, md5=self.annot_md5s[1]
+            )
+            if not self.train:
+                download_url(
+                    url=self.annot_urls[0],
+                    filename="cars_test_annos_withlabels.mat",
+                    root=self.root,
+                    md5=self.annot_md5s[0],
+                )
+
+    def _check_exists(self) -> bool:
+        return (
+            os.path.exists(os.path.join(self.root, f"cars_{'train' if self.train else 'test'}"))
+            and os.path.isdir(os.path.join(self.root, f"cars_{'train' if self.train else 'test'}"))
+            and os.path.exists(os.path.join(self.root, "devkit/cars_meta.mat"))
+            if self.train
+            else os.path.exists(os.path.join(self.root, "cars_test_annos_withlabels.mat"))
+        )