Merge branch 'main' into main

Author: jdsgomes

docs/source/datasets.rst

@@ -70,6 +70,7 @@ You can also create your own datasets using the provided :ref:`base classes <bas
     PCAM
     PhotoTour
     Places365
+    RenderedSST2
     QMNIST
     SBDataset
     SBU

mypy.ini

@@ -121,3 +121,7 @@ ignore_missing_imports = True
 [mypy-torchdata.*]
 
 ignore_missing_imports = True
+
+[mypy-h5py.*]
+
+ignore_missing_imports = True

test/test_datasets.py

@@ -2281,11 +2281,6 @@ def inject_fake_data(self, tmpdir: str, config):
 class SUN397TestCase(datasets_utils.ImageDatasetTestCase):
     DATASET_CLASS = datasets.SUN397
 
-    ADDITIONAL_CONFIGS = datasets_utils.combinations_grid(
-        split=("train", "test"),
-        partition=(1, 10, None),
-    )
-
     def inject_fake_data(self, tmpdir: str, config):
         data_dir = pathlib.Path(tmpdir) / "SUN397"
         data_dir.mkdir()
@@ -2308,18 +2303,7 @@ def inject_fake_data(self, tmpdir: str, config):
         with open(data_dir / "ClassName.txt", "w") as file:
             file.writelines("\n".join(f"/{cls[0]}/{cls}" for cls in sampled_classes))
 
-        if config["partition"] is not None:
-            num_samples = max(len(im_paths) // (2 if config["split"] == "train" else 3), 1)
-
-            with open(data_dir / f"{config['split'].title()}ing_{config['partition']:02d}.txt", "w") as file:
-                file.writelines(
-                    "\n".join(
-                        f"/{f_path.relative_to(data_dir).as_posix()}"
-                        for f_path in random.choices(im_paths, k=num_samples)
-                    )
-                )
-        else:
-            num_samples = len(im_paths)
+        num_samples = len(im_paths)
 
         return num_samples
 
@@ -2397,17 +2381,17 @@ class GTSRBTestCase(datasets_utils.ImageDatasetTestCase):
     DATASET_CLASS = datasets.GTSRB
     FEATURE_TYPES = (PIL.Image.Image, int)
 
-    ADDITIONAL_CONFIGS = datasets_utils.combinations_grid(train=(True, False))
+    ADDITIONAL_CONFIGS = datasets_utils.combinations_grid(split=("train", "test"))
 
     def inject_fake_data(self, tmpdir: str, config):
-        root_folder = os.path.join(tmpdir, "GTSRB")
+        root_folder = os.path.join(tmpdir, "gtsrb")
         os.makedirs(root_folder, exist_ok=True)
 
         # Train data
-        train_folder = os.path.join(root_folder, "Training")
+        train_folder = os.path.join(root_folder, "GTSRB", "Training")
         os.makedirs(train_folder, exist_ok=True)
 
-        num_examples = 3
+        num_examples = 3 if config["split"] == "train" else 4
         classes = ("00000", "00042", "00012")
         for class_idx in classes:
             datasets_utils.create_image_folder(
@@ -2419,7 +2403,7 @@ def inject_fake_data(self, tmpdir: str, config):
 
         total_number_of_examples = num_examples * len(classes)
         # Test data
-        test_folder = os.path.join(root_folder, "Final_Test", "Images")
+        test_folder = os.path.join(root_folder, "GTSRB", "Final_Test", "Images")
         os.makedirs(test_folder, exist_ok=True)
 
         with open(os.path.join(root_folder, "GT-final_test.csv"), "w") as csv_file:
@@ -2665,5 +2649,27 @@ def inject_fake_data(self, tmpdir: str, config):
         return num_images
 
 
+class RenderedSST2TestCase(datasets_utils.ImageDatasetTestCase):
+    DATASET_CLASS = datasets.RenderedSST2
+    ADDITIONAL_CONFIGS = datasets_utils.combinations_grid(split=("train", "val", "test"))
+    SPLIT_TO_FOLDER = {"train": "train", "val": "valid", "test": "test"}
+
+    def inject_fake_data(self, tmpdir: str, config):
+        root_folder = pathlib.Path(tmpdir) / "rendered-sst2"
+        image_folder = root_folder / self.SPLIT_TO_FOLDER[config["split"]]
+
+        num_images_per_class = {"train": 5, "test": 6, "val": 7}
+        sampled_classes = ["positive", "negative"]
+        for cls in sampled_classes:
+            datasets_utils.create_image_folder(
+                image_folder,
+                cls,
+                file_name_fn=lambda idx: f"{idx}.png",
+                num_examples=num_images_per_class[config["split"]],
+            )
+
+        return len(sampled_classes) * num_images_per_class[config["split"]]
+
+
 if __name__ == "__main__":
     unittest.main()

test/test_prototype_datasets_api.py

@@ -126,13 +126,16 @@ def test_default_config(self, info):
         assert info.default_config == default_config
 
     @pytest.mark.parametrize(
-        ("options", "expected_error_msg"),
+        ("valid_options", "options", "expected_error_msg"),
         [
-            pytest.param(dict(unknown_option=None), "Unknown option 'unknown_option'", id="unknown_option"),
-            pytest.param(dict(split="unknown_split"), "Invalid argument 'unknown_split'", id="invalid_argument"),
+            (dict(), dict(any_option=None), "does not take any options"),
+            (dict(split="train"), dict(unknown_option=None), "Unknown option 'unknown_option'"),
+            (dict(split="train"), dict(split="invalid_argument"), "Invalid argument 'invalid_argument'"),
         ],
     )
-    def test_make_config_invalid_inputs(self, info, options, expected_error_msg):
+    def test_make_config_invalid_inputs(self, info, valid_options, options, expected_error_msg):
+        info = make_minimal_dataset_info(valid_options=valid_options)
+
         with pytest.raises(ValueError, match=expected_error_msg):
             info.make_config(**options)
 

test/test_prototype_models.py

@@ -97,7 +97,7 @@ def test_naming_conventions(model_fn):
 )
 @run_if_test_with_prototype
 def test_schema_meta_validation(model_fn):
-    classification_fields = ["size", "categories", "acc@1", "acc@5"]
+    classification_fields = ["size", "categories", "acc@1", "acc@5", "min_size"]
     defaults = {
         "all": ["task", "architecture", "publication_year", "interpolation", "recipe", "num_params"],
         "models": classification_fields,

torchvision/datasets/__init__.py

@@ -29,6 +29,7 @@
 from .pcam import PCAM
 from .phototour import PhotoTour
 from .places365 import Places365
+from .rendered_sst2 import RenderedSST2
 from .sbd import SBDataset
 from .sbu import SBU
 from .semeion import SEMEION
@@ -102,4 +103,5 @@
     "Country211",
     "FGVCAircraft",
     "EuroSAT",
+    "RenderedSST2",
 )

torchvision/datasets/clevr.py

@@ -34,7 +34,7 @@ def __init__(
         split: str = "train",
         transform: Optional[Callable] = None,
         target_transform: Optional[Callable] = None,
-        download: bool = True,
+        download: bool = False,
     ) -> None:
         self._split = verify_str_arg(split, "split", ("train", "val", "test"))
         super().__init__(root, transform=transform, target_transform=target_transform)

torchvision/datasets/country211.py

@@ -32,7 +32,7 @@ def __init__(
         split: str = "train",
         transform: Optional[Callable] = None,
         target_transform: Optional[Callable] = None,
-        download: bool = True,
+        download: bool = False,
     ) -> None:
         self._split = verify_str_arg(split, "split", ("train", "valid", "test"))
 

torchvision/datasets/dtd.py

@@ -21,12 +21,12 @@ class DTD(VisionDataset):
                 The partition only changes which split each image belongs to. Thus, regardless of the selected
                 partition, combining all splits will result in all images.
 
-        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.
         transform (callable, optional): A function/transform that  takes in a 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. Default is False.
     """
 
     _URL = "https://www.robots.ox.ac.uk/~vgg/data/dtd/download/dtd-r1.0.1.tar.gz"
@@ -37,9 +37,9 @@ def __init__(
         root: str,
         split: str = "train",
         partition: int = 1,
-        download: bool = True,
         transform: Optional[Callable] = None,
         target_transform: Optional[Callable] = None,
+        download: bool = False,
     ) -> None:
         self._split = verify_str_arg(split, "split", ("train", "val", "test"))
         if not isinstance(partition, int) and not (1 <= partition <= 10):

torchvision/datasets/eurosat.py

@@ -1,5 +1,5 @@
 import os
-from typing import Any
+from typing import Callable, Optional
 
 from .folder import ImageFolder
 from .utils import download_and_extract_archive
@@ -10,23 +10,21 @@ class EuroSAT(ImageFolder):
 
     Args:
         root (string): Root directory of dataset where ``root/eurosat`` exists.
-        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. Default is False.
         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. Default is False.
     """
 
-    url = "https://madm.dfki.de/files/sentinel/EuroSAT.zip"
-    md5 = "c8fa014336c82ac7804f0398fcb19387"
-
     def __init__(
         self,
         root: str,
+        transform: Optional[Callable] = None,
+        target_transform: Optional[Callable] = None,
         download: bool = False,
-        **kwargs: Any,
     ) -> None:
         self.root = os.path.expanduser(root)
         self._base_folder = os.path.join(self.root, "eurosat")
@@ -38,7 +36,7 @@ def __init__(
         if not self._check_exists():
             raise RuntimeError("Dataset not found. You can use download=True to download it")
 
-        super().__init__(self._data_folder, **kwargs)
+        super().__init__(self._data_folder, transform=transform, target_transform=target_transform)
         self.root = os.path.expanduser(root)
 
     def __len__(self) -> int:
@@ -53,4 +51,8 @@ def download(self) -> None:
             return
 
         os.makedirs(self._base_folder, exist_ok=True)
-        download_and_extract_archive(self.url, download_root=self._base_folder, md5=self.md5)
+        download_and_extract_archive(
+            "https://madm.dfki.de/files/sentinel/EuroSAT.zip",
+            download_root=self._base_folder,
+            md5="c8fa014336c82ac7804f0398fcb19387",
+        )