Add GlaS dataset

scripts/datasets/histopathology/check_glas.py

@@ -0,0 +1,27 @@
+import os
+import sys
+
+from torch_em.util.debug import check_loader
+from torch_em.data.datasets import get_glas_loader
+
+
+sys.path.append("..")
+
+
+def check_glas():
+    from util import ROOT
+
+    loader = get_glas_loader(
+        path=os.path.join(ROOT, "glas"),
+        batch_size=2,
+        patch_shape=(512, 512),
+        split="train",
+        resize_inputs=True,
+        download=True,
+    )
+
+    check_loader(loader, 8, rgb=True)
+
+
+if __name__ == "__main__":
+    check_glas()

torch_em/data/datasets/histopathology/__init__.py

@@ -2,6 +2,7 @@
 from .consep import get_consep_loader, get_consep_dataset
 from .cpm import get_cpm_loader, get_cpm_dataset
 from .cryonuseg import get_cryonuseg_loader, get_cryonuseg_dataset
+from .glas import get_glas_loader, get_glas_dataset
 from .janowczyk import get_janowczyk_loader, get_janowczyk_dataset
 from .lizard import get_lizard_loader, get_lizard_dataset
 from .lynsec import get_lynsec_loader, get_lynsec_dataset

torch_em/data/datasets/histopathology/glas.py

@@ -0,0 +1,172 @@
+"""The GlaS dataset contains annotations for gland segmentation in colon histology images.
+
+This dataset is located at https://www.kaggle.com/datasets/sani84/glasmiccai2015-gland-segmentation.
+The dataset is from the publication http://arxiv.org/abs/1603.00275.
+Please cite it if you use this dataset for your research.
+"""
+
+import os
+import shutil
+from glob import glob
+from tqdm import tqdm
+from natsort import natsorted
+from typing import Union, Tuple, List, Literal
+
+import imageio.v3 as imageio
+
+import torch_em
+
+from torch.utils.data import Dataset, DataLoader
+
+from .. import util
+
+
+def _extract_images(split, path):
+    import h5py
+
+    data_folder = os.path.join(path, "Warwick_QU_Dataset")
+
+    label_paths = natsorted(glob(os.path.join(data_folder, f"{split}*anno.bmp")))
+    image_paths = [
+        image_path for image_path in natsorted(glob(os.path.join(data_folder, f"{split}*.bmp")))
+        if image_path not in label_paths
+    ]
+    assert image_paths and len(image_paths) == len(label_paths)
+
+    os.makedirs(os.path.join(path, split), exist_ok=True)
+
+    for image_path, label_path in tqdm(
+        zip(image_paths, label_paths), total=len(image_paths),
+        desc=f"Extract images from {os.path.abspath(data_folder)}"
+    ):
+        fname = os.path.basename(image_path).split(".")[0]
+
+        image = imageio.imread(image_path)
+        segmentation = imageio.imread(label_path)
+        image = image.transpose(2, 0, 1)
+
+        with h5py.File(os.path.join(path, split, f"{fname}.h5"), "a") as f:
+            f.create_dataset("raw", data=image, compression="gzip")
+            f.create_dataset("labels", data=segmentation, compression="gzip")
+
+
+def get_glas_data(
+    path: Union[os.PathLike, str], split: Literal["train", "val", "test"], download: bool = False
+) -> str:
+    """Download the GlaS dataset.
+
+    Args:
+        path: Filepath to a folder where the data is downloaded for further processing.
+        split: The choice of data split.
+        download: Whether to download the data if it is not present.
+
+    Returns:
+        Filepath where the data is downloaded and preprocessed.
+    """
+    data_dir = os.path.join(path, split)
+    if os.path.exists(data_dir):
+        return data_dir
+
+    os.makedirs(path, exist_ok=True)
+
+    # Download the files.
+    util.download_source_kaggle(path=path, dataset_name="sani84/glasmiccai2015-gland-segmentation", download=download)
+    util.unzip(zip_path=os.path.join(path, "glasmiccai2015-gland-segmentation.zip"), dst=path)
+
+    # Preprocess inputs per split.
+    splits = ["train", "test"]
+    if split not in splits:
+        raise ValueError(f"'{split}' is not a valid split choice.")
+
+    for _split in splits:
+        _extract_images(_split, path)
+
+    # Remove original data
+    shutil.rmtree(os.path.join(path, "Warwick_QU_Dataset"))
+
+    return data_dir
+
+
+def get_glas_paths(path: Union[os.PathLike], split: Literal["train", "test"], download: bool = False) -> List[str]:
+    """Get paths to the GlaS data.
+
+    Args:
+        path: Filepath to a folder where the downloaded data will be saved.
+        split: The choice of data split.
+        download: Whether to download the data if it is not present.
+
+    Returns:
+        List of filepaths for the stored data.
+    """
+    data_dir = get_glas_data(path, split, download)
+    data_paths = natsorted(glob(os.path.join(data_dir, "*.h5")))
+    return data_paths
+
+
+def get_glas_dataset(
+    path: Union[os.PathLike, str],
+    patch_shape: Tuple[int, int],
+    split: Literal["train", "test"],
+    resize_inputs: bool = False,
+    download: bool = False,
+    **kwargs
+) -> Dataset:
+    """Get the GlaS dataset for gland segmentation.
+
+    Args:
+        path: Filepath to a folder where the downloaded data will be saved.
+        patch_shape: The patch shape to use for training.
+        split: The choice of data split.
+        resize_inputs: Whether to resize the input images.
+        download: Whether to download the data if it is not present.
+        kwargs: Additional keyword arguments for `torch_em.default_segmentation_dataset`.
+
+    Returns:
+        The segmentation dataset.
+    """
+    data_paths = get_glas_paths(path, split, download)
+
+    if resize_inputs:
+        resize_kwargs = {"patch_shape": patch_shape, "is_rgb": True}
+        kwargs, patch_shape = util.update_kwargs_for_resize_trafo(
+            kwargs=kwargs, patch_shape=patch_shape, resize_inputs=resize_inputs, resize_kwargs=resize_kwargs
+        )
+
+    return torch_em.default_segmentation_dataset(
+        raw_paths=data_paths,
+        raw_key="raw",
+        label_paths=data_paths,
+        label_key="labels",
+        patch_shape=patch_shape,
+        ndim=2,
+        with_channels=True,
+        **kwargs
+    )
+
+
+def get_glas_loader(
+    path: Union[os.PathLike, str],
+    batch_size: int,
+    patch_shape: Tuple[int, int],
+    split: Literal["train", "test"],
+    resize_inputs: bool = False,
+    download: bool = False,
+    **kwargs
+) -> DataLoader:
+    """Get the GlaS dataloader for gland segmentation.
+
+    Args:
+        path: Filepath to a folder where the downloaded data will be saved.
+        batch_size: The batch size for training.
+        patch_shape: The patch shape to use for training.
+        split: The choice of data split.
+        resize_inputs: Whether to resize the inputs.
+        download: Whether to download the data if it is not present.
+        kwargs: Additional keyword arguments for `torch_em.default_segmentation_dataset` or for the PyTorch DataLoader.
+
+    Returns:
+        The DataLoader.
+    """
+    ds_kwargs, loader_kwargs = util.split_kwargs(torch_em.default_segmentation_dataset, **kwargs)
+    dataset = get_glas_dataset(path, patch_shape, split, resize_inputs, download, **ds_kwargs)
+    return torch_em.get_data_loader(dataset, batch_size, **loader_kwargs)