When creating processing instances some arguments may be not Tuple bu…

…t ListConfig and silently serialized to state dict with pickle. By adding explicit casting to primitive tuple type we fix that (#1534)

src/super_gradients/training/processing/processing.py

@@ -139,7 +139,7 @@ class ImagePermute(Processing):
     """
 
     def __init__(self, permutation: Tuple[int, int, int] = (2, 0, 1)):
-        self.permutation = permutation
+        self.permutation = tuple(permutation)
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, None]:
         processed_image = np.ascontiguousarray(image.transpose(*self.permutation))
@@ -187,7 +187,7 @@ class StandardizeImage(Processing):
 
     def __init__(self, max_value: float = 255.0):
         super().__init__()
-        self.max_value = max_value
+        self.max_value = float(max_value)
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, None]:
         """Reverse the channel order of an image.
@@ -246,7 +246,7 @@ class _DetectionPadding(Processing, ABC):
     """
 
     def __init__(self, output_shape: Tuple[int, int], pad_value: int):
-        self.output_shape = output_shape
+        self.output_shape = tuple(output_shape)
         self.pad_value = pad_value
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, DetectionPadToSizeMetadata]:
@@ -288,7 +288,7 @@ class _KeypointsPadding(Processing, ABC):
     """
 
     def __init__(self, output_shape: Tuple[int, int], pad_value: int):
-        self.output_shape = output_shape
+        self.output_shape = tuple(output_shape)
         self.pad_value = pad_value
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, DetectionPadToSizeMetadata]:
@@ -351,7 +351,7 @@ class _Rescale(Processing, ABC):
     """
 
     def __init__(self, output_shape: Tuple[int, int]):
-        self.output_shape = output_shape
+        self.output_shape = tuple(output_shape)
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, RescaleMetadata]:
         scale_factor_h, scale_factor_w = self.output_shape[0] / image.shape[0], self.output_shape[1] / image.shape[1]
@@ -378,7 +378,7 @@ class _LongestMaxSizeRescale(Processing, ABC):
     """
 
     def __init__(self, output_shape: Tuple[int, int]):
-        self.output_shape = output_shape
+        self.output_shape = tuple(output_shape)
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, RescaleMetadata]:
         height, width = image.shape[:2]
@@ -443,7 +443,7 @@ def postprocess_predictions(self, predictions: Prediction, metadata: None) -> Pr
 class Resize(ClassificationProcess):
     def __init__(self, size: int = 224):
         super().__init__()
-        self.size = size
+        self.size = int(size)
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, None]:
         """Resize an image.
@@ -477,7 +477,7 @@ class CenterCrop(ClassificationProcess):
 
     def __init__(self, size: int = 224):
         super().__init__()
-        self.size = size
+        self.size = int(size)
 
     def preprocess_image(self, image: np.ndarray) -> Tuple[np.ndarray, None]:
         """Crops the given image at the center.