diff --git a/src/matchcake_opt/datasets/__init__.py b/src/matchcake_opt/datasets/__init__.py
index adf6066..5f2803f 100644
--- a/src/matchcake_opt/datasets/__init__.py
+++ b/src/matchcake_opt/datasets/__init__.py
@@ -6,6 +6,7 @@
 from .digits2d import Digits2D
 from .mnist_dataset import MNISTDataset
 from .pathmnist_dataset import PathMNISTDataset
+from .retinamnist_dataset import RetinaMNISTDataset
 
 dataset_name_to_type_map: Dict[str, Type[BaseDataset]] = {
     _cls.DATASET_NAME: _cls
diff --git a/src/matchcake_opt/datasets/retinamnist_dataset.py b/src/matchcake_opt/datasets/retinamnist_dataset.py
new file mode 100644
index 0000000..8d89195
--- /dev/null
+++ b/src/matchcake_opt/datasets/retinamnist_dataset.py
@@ -0,0 +1,56 @@
+from pathlib import Path
+from typing import Sequence, Union
+
+import numpy as np
+import torch
+from medmnist import RetinaMNIST
+from torch.utils.data import ConcatDataset
+from torchvision import transforms
+from torchvision.transforms import ToTensor
+from torchvision.transforms.v2 import Compose
+
+from .base_dataset import BaseDataset
+
+
+class RetinaMNISTDataset(BaseDataset):
+    DATASET_NAME = "RetinaMNIST"
+
+    @staticmethod
+    def to_scalar_tensor(y: Sequence[int]):
+        return torch.tensor(y).item()
+
+    @staticmethod
+    def to_long_tensor(y: Sequence[int]) -> torch.Tensor:
+        return torch.tensor(y, dtype=torch.long)
+
+    def __init__(self, data_dir: Union[str, Path] = Path("./data/") / DATASET_NAME, train: bool = True, **kwargs):
+        super().__init__(data_dir, train, **kwargs)
+        transform = Compose(
+            [
+                ToTensor(),
+                transforms.Normalize(0.0, 1.0),
+            ]
+        )
+        target_transform = Compose([self.to_scalar_tensor, self.to_long_tensor])
+        self._data = RetinaMNIST(
+            root=self.data_dir,
+            split="train" if self.train else "test",
+            download=True,
+            transform=transform,
+            target_transform=target_transform,
+        )
+        self._n_classes = np.unique(self._data.labels).size
+        if self.train:
+            val_dataset = RetinaMNIST(
+                root=self.data_dir, split="val", download=True, transform=transform, target_transform=target_transform
+            )
+            self._data = ConcatDataset([self._data, val_dataset])
+
+    def __getitem__(self, item):
+        return self._data[item]
+
+    def __len__(self):
+        return len(self._data)
+
+    def get_output_shape(self) -> tuple:
+        return (self._n_classes,)
diff --git a/tests/test_datasets/test_retinamnist_dataset.py b/tests/test_datasets/test_retinamnist_dataset.py
new file mode 100644
index 0000000..9dbe159
--- /dev/null
+++ b/tests/test_datasets/test_retinamnist_dataset.py
@@ -0,0 +1,63 @@
+import shutil
+from pathlib import Path
+from unittest.mock import MagicMock
+
+import numpy as np
+import pytest
+import torch
+
+from matchcake_opt.datasets.retinamnist_dataset import RetinaMNISTDataset
+
+
+class TestRetinaMNISTDataset:
+    MOCK_LEN = 10
+
+    @pytest.fixture(scope="class")
+    def data_dir(self):
+        path = Path(".tmp") / "data_dir" / "retinamnist"
+        yield path
+        shutil.rmtree(path, ignore_errors=True)
+
+    @pytest.fixture
+    def data_mock(self, monkeypatch):
+        cls_mock = MagicMock()
+        monkeypatch.setattr("matchcake_opt.datasets.retinamnist_dataset.RetinaMNIST", cls_mock)
+        mock = MagicMock()
+        cls_mock.return_value = mock
+        mock.__getitem__.return_value = (torch.zeros(3, 28, 28), torch.zeros(1).long())
+        mock.__len__.return_value = self.MOCK_LEN
+        mock.labels = np.arange(5)
+        monkeypatch.setattr("matchcake_opt.datasets.retinamnist_dataset.ConcatDataset", lambda *x: mock)
+        return mock
+
+    @pytest.fixture
+    def dataset_instance(self, data_mock, data_dir):
+        return RetinaMNISTDataset(data_dir=data_dir, train=True)
+
+    def test_init(self, data_mock, data_dir):
+        dataset = RetinaMNISTDataset(data_dir=data_dir, train=True)
+        assert dataset._data == data_mock
+
+    def test_getitem(self, data_mock, dataset_instance):
+        datum = dataset_instance[0]
+        assert isinstance(datum, tuple)
+        assert isinstance(datum[0], torch.Tensor)
+        assert isinstance(datum[1], torch.Tensor)
+        assert datum[1].dtype == torch.long
+        data_mock.__getitem__.assert_called_once_with(0)
+
+    def test_to_long_tensor(self, dataset_instance):
+        x = torch.zeros(0, dtype=torch.int)
+        y = dataset_instance.to_long_tensor(x)
+        assert y.dtype == torch.long
+
+    def test_len(self, dataset_instance):
+        assert len(dataset_instance) == self.MOCK_LEN
+
+    def test_output_shape(self, dataset_instance):
+        assert dataset_instance.get_output_shape() == (5,)
+
+    def test_to_scalar_tensor(self, dataset_instance):
+        x = torch.tensor([1])
+        y = dataset_instance.to_scalar_tensor(x)
+        assert isinstance(y, int)