add unit test

src/otx/core/data/transform_libs/torchvision.py

@@ -635,7 +635,7 @@ class RandomResizedCrop(tvt_v2.Transform, NumpytoTVTensorMixin):
     is made. This crop is finally resized to given size.
 
     Args:
-        scale (sequence | int): Desired output scale of the crop. If size is an
+        scale (Sequence[int] | int): Desired output scale of the crop. If size is an
             int instead of sequence like (h, w), a square crop (size, size) is
             made.
         crop_ratio_range (tuple): Range of the random size of the cropped
@@ -654,7 +654,7 @@ class RandomResizedCrop(tvt_v2.Transform, NumpytoTVTensorMixin):
 
     def __init__(
         self,
-        scale: Sequence | int,
+        scale: Sequence[int] | int,
         crop_ratio_range: tuple[float, float] = (0.08, 1.0),
         aspect_ratio_range: tuple[float, float] = (3.0 / 4.0, 4.0 / 3.0),
         max_attempts: int = 10,
@@ -2342,7 +2342,7 @@ class RandomCrop(tvt_v2.Transform, NumpytoTVTensorMixin):
     The absolute `crop_size` is sampled based on `crop_type` and `image_size`, then the cropped results are generated.
 
     Args:
-        crop_size (tuple): The relative ratio or absolute pixels of
+        crop_size (tuple[int, int]): The relative ratio or absolute pixels of
             (height, width).
         crop_type (str, optional): One of "relative_range", "relative",
             "absolute", "absolute_range". "relative" randomly crops
@@ -2367,7 +2367,7 @@ class RandomCrop(tvt_v2.Transform, NumpytoTVTensorMixin):
 
     def __init__(
         self,
-        crop_size: tuple,  # (H, W)
+        crop_size: tuple[int, int],  # (H, W)
         crop_type: str = "absolute",
         cat_max_ratio: int | float = 1,
         allow_negative_crop: bool = False,
@@ -3122,9 +3122,7 @@ def generate(cls, config: SubsetConfig) -> Compose:
 
         transforms = []
         for cfg_transform in config.transforms:
-            for val in cfg_transform["init_args"].values():
-                if isinstance(val, str) and "^{input_size}" in val:
-                    cls._eval_input_size(cfg_transform, config.input_size)
+            cls._eval_input_size(cfg_transform, config.input_size)
             transform = cls._dispatch_transform(cfg_transform)
             transforms.append(transform)
 
@@ -3143,6 +3141,8 @@ def _eval_input_size(cls, cfg_transform: dict[str, Any], input_size: int | tuple
             return
         if isinstance(input_size, int):
             input_size = (input_size, input_size)
+        else:
+            input_size = tuple(input_size)
 
         def check_type(value, expected_type) -> bool:
             try:
@@ -3160,9 +3160,11 @@ def check_type(value, expected_type) -> bool:
 
             available_types = typing.get_type_hints(model_cls.__init__).get(key)
             if available_types is None or check_type(input_size, available_types):  # pass tuple[int, int]
-                cfg_transform["init_args"][key] = eval(val.replace("^{input_size}", "np.array(input_size)")).tolist()
+                cfg_transform["init_args"][key] = tuple(
+                    eval(val.replace("^{input_size}", "np.array(input_size)")).round().astype(np.int32).tolist()
+                )
             elif check_type(input_size[0], available_types):  # pass int
-                cfg_transform["init_args"][key] = eval(val.replace("^{input_size}", "input_size[0]"))
+                cfg_transform["init_args"][key] = round(eval(val.replace("^{input_size}", "input_size[0]")))
             else:
                 msg = f"{key} argument should be able to get int or tuple[int, int], but it can get {available_types}"
                 raise RuntimeError(msg)

tests/unit/core/data/test_module.py

@@ -46,18 +46,21 @@ def fxt_config(self) -> DictConfig:
         )
         train_subset.num_workers = 0
         train_subset.batch_size = 4
+        train_subset.input_size = None
         val_subset = MagicMock(spec=SubsetConfig)
         val_subset.sampler = DictConfig(
             {"class_path": "torch.utils.data.RandomSampler", "init_args": {"num_samples": 3}},
         )
         val_subset.num_workers = 0
         val_subset.batch_size = 3
+        val_subset.input_size = None
         test_subset = MagicMock(spec=SubsetConfig)
         test_subset.sampler = DictConfig(
             {"class_path": "torch.utils.data.RandomSampler", "init_args": {"num_samples": 3}},
         )
         test_subset.num_workers = 0
         test_subset.batch_size = 1
+        test_subset.input_size = None
         unlabeled_subset = MagicMock(spec=UnlabeledDataConfig)
         unlabeled_subset.data_root = None
         tile_config = MagicMock(spec=TileConfig)
@@ -101,6 +104,7 @@ def test_init(
         fxt_config.train_subset.subset_name = "train_1"
         fxt_config.val_subset.subset_name = "val_1"
         fxt_config.test_subset.subset_name = "test_1"
+        input_size = (512, 512)
 
         # Dataset will have "train_0", "train_1", "val_0", ..., "test_1" subsets
         mock_dm_subsets = {f"{name}_{idx}": MagicMock() for name in ["train", "val", "test"] for idx in range(2)}
@@ -115,13 +119,17 @@ def test_init(
             train_subset=fxt_config.train_subset,
             val_subset=fxt_config.val_subset,
             test_subset=fxt_config.test_subset,
+            input_size=input_size,
         )
 
         assert module.train_dataloader().batch_size == 4
         assert module.val_dataloader().batch_size == 3
         assert module.test_dataloader().batch_size == 1
         assert module.predict_dataloader().batch_size == 1
         assert mock_otx_dataset_factory.create.call_count == 3
+        assert fxt_config.train_subset.input_size == input_size
+        assert fxt_config.val_subset.input_size == input_size
+        assert fxt_config.test_subset.input_size == input_size
 
     @patch("otx.core.data.module.OTXDatasetFactory")
     @patch("otx.core.data.module.DmDataset.import_from")

tests/unit/core/data/test_transform_libs.py

@@ -186,6 +186,36 @@ def test_transform(
         item = dataset[0]
         assert isinstance(item, data_entity_cls)
 
+    @pytest.fixture
+    def fxt_config_w_input_size(self) -> list[dict[str, Any]]:
+        cfg = f"""
+        input_size:
+        - 224
+        - 224
+        transforms:
+          - class_path: otx.core.data.transform_libs.torchvision.ResizetoLongestEdge
+            init_args:
+                size: ^{{input_size}} * 2
+          - class_path: otx.core.data.transform_libs.torchvision.RandomResize
+            init_args:
+                scale: ^{{input_size}} * 0.5
+          - class_path: otx.core.data.transform_libs.torchvision.RandomCrop
+            init_args:
+                crop_size: ^{{input_size}}
+          - class_path: otx.core.data.transform_libs.torchvision.RandomResize
+            init_args:
+                scale: ^{{input_size}} * 1.1
+        """
+        return OmegaConf.create(cfg)
+
+    def test_eval_input_size(self, fxt_config_w_input_size):
+        transform = TorchVisionTransformLib.generate(fxt_config_w_input_size)
+        assert isinstance(transform, v2.Compose)
+        assert transform.transforms[0].size == 448  # ResizetoLongestEdge gets an integer
+        assert transform.transforms[1].scale == (112, 112)  # RandomResize gets sequence of integer
+        assert transform.transforms[2].crop_size == (224, 224)  # RandomCrop gets sequence of integer
+        assert transform.transforms[3].scale == (round(224 * 1.1), round(224 * 1.1))  # check round
+
     @pytest.fixture(params=["RGB", "BGR"])
     def fxt_image_color_channel(self, request) -> ImageColorChannel:
         return ImageColorChannel(request.param)

tests/unit/core/utils/test_utils.py

@@ -11,6 +11,7 @@
     get_mean_std_from_data_processing,
     is_ckpt_for_finetuning,
     is_ckpt_from_otx_v1,
+    get_obj_from_str,
 )
 
 
@@ -113,3 +114,8 @@ def test_get_idx_list_per_classes(fxt_dm_dataset):
     expected_result["0"] = list(range(100))
     expected_result["1"] = list(range(100, 108))
     assert result == expected_result
+
+def test_get_obj_from_str():
+    obj_path = "otx.core.utils.utils.get_mean_std_from_data_processing"
+    obj = get_obj_from_str(obj_path)
+    assert obj == get_mean_std_from_data_processing