✨ Define SemanticSegmentor with the New EngineABC

tests/conftest.py

@@ -520,6 +520,7 @@ def sample_wsi_dict(remote_sample: Callable) -> dict:
         "wsi4_4k_4k_svs",
         "wsi3_20k_20k_pred",
         "wsi4_4k_4k_pred",
+        "wsi4_1k_1k_svs",
     ]
     return {name: remote_sample(name) for name in file_names}
 

tests/engines/test_engine_abc.py

@@ -520,7 +520,7 @@ def test_io_config_delegation(tmp_path: Path, caplog: pytest.LogCaptureFixture)
     eng = TestEngineABC(model=model)
 
     kwargs = {
-        "patch_input_shape": [512, 512],
+        "patch_input_shape": [224, 224],
         "input_resolutions": [{"units": "mpp", "resolution": 1.75}],
     }
     with caplog.at_level(logging.WARNING):
@@ -536,7 +536,7 @@ def test_io_config_delegation(tmp_path: Path, caplog: pytest.LogCaptureFixture)
 
     # test providing config / full input info for non pretrained models
     ioconfig = ModelIOConfigABC(
-        patch_input_shape=(512, 512),
+        patch_input_shape=(224, 224),
         stride_shape=(256, 256),
         input_resolutions=[{"resolution": 1.35, "units": "mpp"}],
     )
@@ -546,7 +546,7 @@ def test_io_config_delegation(tmp_path: Path, caplog: pytest.LogCaptureFixture)
         save_dir=f"{tmp_path}/dump",
         ioconfig=ioconfig,
     )
-    assert eng._ioconfig.patch_input_shape == (512, 512)
+    assert eng._ioconfig.patch_input_shape == (224, 224)
     assert eng._ioconfig.stride_shape == (256, 256)
     assert eng._ioconfig.input_resolutions == [{"resolution": 1.35, "units": "mpp"}]
     shutil.rmtree(tmp_path / "dump", ignore_errors=True)
@@ -557,15 +557,15 @@ def test_io_config_delegation(tmp_path: Path, caplog: pytest.LogCaptureFixture)
         save_dir=f"{tmp_path}/dump",
         **kwargs,
     )
-    assert eng._ioconfig.patch_input_shape == [512, 512]
-    assert eng._ioconfig.stride_shape == [512, 512]
+    assert eng._ioconfig.patch_input_shape == [224, 224]
+    assert eng._ioconfig.stride_shape == [224, 224]
     assert eng._ioconfig.input_resolutions == [{"resolution": 1.75, "units": "mpp"}]
     shutil.rmtree(tmp_path / "dump", ignore_errors=True)
 
     # test overwriting pretrained ioconfig
     eng = TestEngineABC(model="alexnet-kather100k")
     eng.run(
-        images=np.zeros((10, 224, 224, 3), dtype=np.uint8),
+        images=np.zeros((10, 300, 300, 3), dtype=np.uint8),
         patch_input_shape=(300, 300),
         stride_shape=(300, 300),
         input_resolutions=[{"units": "baseline", "resolution": 1.99}],
@@ -579,7 +579,7 @@ def test_io_config_delegation(tmp_path: Path, caplog: pytest.LogCaptureFixture)
     shutil.rmtree(tmp_path / "dump", ignore_errors=True)
 
     eng.run(
-        images=np.zeros((10, 224, 224, 3), dtype=np.uint8),
+        images=np.zeros((10, 300, 300, 3), dtype=np.uint8),
         patch_input_shape=(300, 300),
         stride_shape=(300, 300),
         input_resolutions=None,

tests/engines/test_semantic_segmentor.py

@@ -0,0 +1,193 @@
+"""Test SemanticSegmentor."""
+
+from __future__ import annotations
+
+import json
+import sqlite3
+from typing import TYPE_CHECKING, Callable
+
+import numpy as np
+import torch
+import zarr
+
+from tiatoolbox.annotation import SQLiteStore
+from tiatoolbox.models.engine.semantic_segmentor_new import SemanticSegmentor
+from tiatoolbox.utils import env_detection as toolbox_env
+from tiatoolbox.utils.misc import imread
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+device = "cuda" if toolbox_env.has_gpu() else "cpu"
+
+
+def test_semantic_segmentor_init() -> None:
+    """Tests SemanticSegmentor initialization."""
+    segmentor = SemanticSegmentor(model="fcn-tissue_mask", device=device)
+
+    assert isinstance(segmentor, SemanticSegmentor)
+    assert isinstance(segmentor.model, torch.nn.Module)
+
+
+def test_semantic_segmentor_patches(remote_sample: Callable, tmp_path: Path) -> None:
+    """Tests SemanticSegmentor on image patches."""
+    segmentor = SemanticSegmentor(
+        model="fcn-tissue_mask", batch_size=32, verbose=False, device=device
+    )
+
+    sample_image = remote_sample("thumbnail-1k-1k")
+
+    inputs = [sample_image, sample_image]
+
+    assert segmentor.cache_mode is False
+
+    output = segmentor.run(
+        images=inputs,
+        return_probabilities=True,
+        return_labels=False,
+        device=device,
+        patch_mode=True,
+    )
+
+    assert 0.15 < np.mean(output["predictions"][:]) < 0.18
+    assert 0.495 < np.mean(output["probabilities"][:]) < 0.505
+
+    assert (
+        tuple(segmentor._ioconfig.patch_output_shape)
+        == output["probabilities"][0].shape[:-1]
+    )
+
+    assert (
+        tuple(segmentor._ioconfig.patch_input_shape) == output["predictions"][0].shape
+    )
+
+    output = segmentor.run(
+        images=inputs,
+        return_probabilities=True,
+        return_labels=False,
+        device=device,
+        patch_mode=True,
+        cache_mode=True,
+        save_dir=tmp_path / "output0",
+    )
+
+    assert output == tmp_path / "output0" / "output.zarr"
+
+    output = zarr.open(output, mode="r")
+    assert 0.15 < np.mean(output["predictions"][:]) < 0.18
+    assert 0.495 < np.mean(output["probabilities"][:]) < 0.505
+
+    output = segmentor.run(
+        images=inputs,
+        return_probabilities=False,
+        return_labels=False,
+        device=device,
+        patch_mode=True,
+        cache_mode=True,
+        output_type="zarr",
+        save_dir=tmp_path / "output1",
+    )
+
+    assert output == tmp_path / "output1" / "output.zarr"
+
+    output = zarr.open(output, mode="r")
+    assert 0.15 < np.mean(output["predictions"][:]) < 0.18
+    assert "probabilities" not in output.keys()  # noqa: SIM118
+
+    output = segmentor.run(
+        images=inputs,
+        return_probabilities=False,
+        return_labels=False,
+        device=device,
+        patch_mode=True,
+        cache_mode=False,
+        save_dir=tmp_path / "output2",
+        output_type="zarr",
+    )
+
+    assert output == tmp_path / "output2" / "output.zarr"
+
+    output = zarr.open(output, mode="r")
+    assert 0.15 < np.mean(output["predictions"][:]) < 0.18
+    assert "probabilities" not in output
+    assert "predictions" in output
+
+
+def _test_store_output_patch(output: Path) -> None:
+    """Helper method to test annotation store output for a patch."""
+    store_ = SQLiteStore.open(output)
+    annotations_ = store_.values()
+    annotations_geometry_type = [
+        str(annotation_.geometry_type) for annotation_ in annotations_
+    ]
+    assert "Polygon" in annotations_geometry_type
+
+    con = sqlite3.connect(output)
+    cur = con.cursor()
+    annotations_properties = list(cur.execute("SELECT properties FROM annotations"))
+
+    out = []
+
+    for item in annotations_properties:
+        for json_str in item:
+            probs = json.loads(json_str)
+            if "type" in probs:
+                out.append(probs.pop("type"))
+
+    assert "mask" in out
+
+    assert annotations_properties is not None
+
+
+def test_save_annotation_store(remote_sample: Callable, tmp_path: Path) -> None:
+    """Test for saving output as annotation store."""
+    segmentor = SemanticSegmentor(
+        model="fcn-tissue_mask", batch_size=32, verbose=False, device=device
+    )
+
+    sample_image = remote_sample("thumbnail-1k-1k")
+
+    inputs = [sample_image]
+
+    output = segmentor.run(
+        images=inputs,
+        return_probabilities=False,
+        return_labels=False,
+        device=device,
+        patch_mode=True,
+        cache_mode=True,
+        save_dir=tmp_path / "output1",
+        output_type="annotationstore",
+    )
+
+    assert output[0] == tmp_path / "output1" / (sample_image.stem + ".db")
+    _test_store_output_patch(output[0])
+
+
+def test_save_annotation_store_nparray(remote_sample: Callable, tmp_path: Path) -> None:
+    """Test for saving output as annotation store using a numpy array."""
+    segmentor = SemanticSegmentor(
+        model="fcn-tissue_mask", batch_size=32, verbose=False, device=device
+    )
+
+    sample_image = remote_sample("thumbnail-1k-1k")
+
+    input_image = imread(sample_image)
+    inputs_list = [input_image, input_image]
+
+    output = segmentor.run(
+        images=inputs_list,
+        return_probabilities=False,
+        return_labels=False,
+        device=device,
+        patch_mode=True,
+        cache_mode=True,
+        save_dir=tmp_path / "output1",
+        output_type="annotationstore",
+    )
+
+    assert output[0] == tmp_path / "output1" / "0.db"
+    assert output[1] == tmp_path / "output1" / "1.db"
+
+    _test_store_output_patch(output[0])
+    _test_store_output_patch(output[1])

tests/models/test_arch_unet.py

@@ -50,7 +50,7 @@ def test_functional_unet(remote_sample: Callable) -> None:
     pretrained = torch.load(pretrained_weights, map_location="cpu")
     model.load_state_dict(pretrained)
     output = model.infer_batch(model, batch, device=select_device(on_gpu=ON_GPU))
-    _ = output[0]
+    _ = output["probabilities"][0]
 
     # run untrained network to test for architecture
     model = UNetModel(

tests/models/test_dataset.py

@@ -20,6 +20,7 @@
 )
 from tiatoolbox.utils import download_data, imread, imwrite, unzip_data
 from tiatoolbox.utils import env_detection as toolbox_env
+from tiatoolbox.utils.exceptions import DimensionMismatchError
 from tiatoolbox.wsicore import WSIReader
 
 RNG = np.random.default_rng()  # Numpy Random Generator
@@ -120,7 +121,9 @@ def test_kather_dataset(tmp_path: Path) -> None:
     assert len(dataset.inputs) == len(dataset.labels)
 
     # to actually get the image, we feed it to PatchDataset
-    actual_ds = PatchDataset(dataset.inputs, dataset.labels)
+    actual_ds = PatchDataset(
+        dataset.inputs, dataset.labels, patch_input_shape=(224, 224)
+    )
     sample_patch = actual_ds[89]
     assert isinstance(sample_patch["image"], np.ndarray)
     assert sample_patch["label"] is not None
@@ -129,14 +132,28 @@ def test_kather_dataset(tmp_path: Path) -> None:
     shutil.rmtree(save_dir_path, ignore_errors=True)
 
 
+def test_incorrect_input_shape() -> None:
+    """Incorrect input patch dimensions should raise DimensionMismatchError."""
+    size = (5, 5, 3)
+    img = RNG.integers(low=0, high=255, size=size)
+    list_imgs = [img, img, img]
+    dataset = PatchDataset(list_imgs, patch_input_shape=(100, 100))
+    with pytest.raises(
+        DimensionMismatchError, match=r".*\(100, 100\), but got \(5, 5\).*"
+    ):
+        _ = dataset[0]
+
+
 def test_patch_dataset_path_imgs(
     sample_patch1: str | Path,
     sample_patch2: str | Path,
 ) -> None:
     """Test for patch dataset with a list of file paths as input."""
     size = (224, 224, 3)
 
-    dataset = PatchDataset([Path(sample_patch1), Path(sample_patch2)])
+    dataset = PatchDataset(
+        [Path(sample_patch1), Path(sample_patch2)], patch_input_shape=size[:-1]
+    )
 
     for _, sample_data in enumerate(dataset):
         sampled_img_shape = sample_data["image"].shape
@@ -152,7 +169,7 @@ def test_patch_dataset_list_imgs(tmp_path: Path) -> None:
     size = (5, 5, 3)
     img = RNG.integers(low=0, high=255, size=size)
     list_imgs = [img, img, img]
-    dataset = PatchDataset(list_imgs)
+    dataset = PatchDataset(list_imgs, patch_input_shape=size[:-1])
 
     dataset.preproc_func = lambda x: x
 
@@ -197,14 +214,14 @@ def test_patch_datasetarray_imgs() -> None:
     array_imgs = np.array(list_imgs)
 
     # test different setter for label
-    dataset = PatchDataset(array_imgs, labels=labels)
+    dataset = PatchDataset(array_imgs, labels=labels, patch_input_shape=(5, 5))
     an_item = dataset[2]
     assert an_item["label"] == 3
-    dataset = PatchDataset(array_imgs, labels=None)
+    dataset = PatchDataset(array_imgs, labels=None, patch_input_shape=(5, 5))
     an_item = dataset[2]
     assert "label" not in an_item
 
-    dataset = PatchDataset(array_imgs)
+    dataset = PatchDataset(array_imgs, patch_input_shape=size[:-1])
     for _, sample_data in enumerate(dataset):
         sampled_img_shape = sample_data["image"].shape
         assert sampled_img_shape[0] == size[0]