diff --git a/src/transformers/models/detr/image_processing_detr_fast.py b/src/transformers/models/detr/image_processing_detr_fast.py
index eadde59e55e4..b414b4224e68 100644
--- a/src/transformers/models/detr/image_processing_detr_fast.py
+++ b/src/transformers/models/detr/image_processing_detr_fast.py
@@ -347,7 +347,7 @@ def __init__(
         format: Union[str, AnnotationFormat] = AnnotationFormat.COCO_DETECTION,
         do_resize: bool = True,
         size: Dict[str, int] = None,
-        resample: [Union[PILImageResampling, F.InterpolationMode]] = PILImageResampling.BILINEAR,
+        resample: Union[PILImageResampling, "F.InterpolationMode"] = PILImageResampling.BILINEAR,
         do_rescale: bool = True,
         rescale_factor: Union[int, float] = 1 / 255,
         do_normalize: bool = True,
@@ -462,7 +462,7 @@ def resize(
         self,
         image: torch.Tensor,
         size: SizeDict,
-        interpolation: F.InterpolationMode = F.InterpolationMode.BILINEAR,
+        interpolation: "F.InterpolationMode" = None,
         **kwargs,
     ) -> torch.Tensor:
         """
@@ -485,6 +485,7 @@ def resize(
             interpolation (`InterpolationMode`, *optional*, defaults to `InterpolationMode.BILINEAR`):
                 Resampling filter to use if resizing the image.
         """
+        interpolation = interpolation if interpolation is not None else F.InterpolationMode.BILINEAR
         if size.shortest_edge and size.longest_edge:
             # Resize the image so that the shortest edge or the longest edge is of the given size
             # while maintaining the aspect ratio of the original image.
@@ -517,7 +518,7 @@ def resize_annotation(
         orig_size: Tuple[int, int],
         target_size: Tuple[int, int],
         threshold: float = 0.5,
-        interpolation: F.InterpolationMode = F.InterpolationMode.NEAREST,
+        interpolation: "F.InterpolationMode" = None,
     ):
         """
         Resizes an annotation to a target size.
@@ -534,6 +535,7 @@ def resize_annotation(
             resample (`InterpolationMode`, defaults to `InterpolationMode.NEAREST`):
                 The resampling filter to use when resizing the masks.
         """
+        interpolation = interpolation if interpolation is not None else F.InterpolationMode.NEAREST
         ratio_height, ratio_width = [target / orig for target, orig in zip(target_size, orig_size)]
 
         new_annotation = {}
@@ -680,7 +682,7 @@ def preprocess(
         masks_path: Optional[Union[str, pathlib.Path]] = None,
         do_resize: Optional[bool] = None,
         size: Optional[Dict[str, int]] = None,
-        resample: Optional[Union[PILImageResampling, F.InterpolationMode]] = None,
+        resample: Optional[Union[PILImageResampling, "F.InterpolationMode"]] = None,
         do_rescale: Optional[bool] = None,
         rescale_factor: Optional[Union[int, float]] = None,
         do_normalize: Optional[bool] = None,
diff --git a/src/transformers/models/rt_detr/image_processing_rt_detr_fast.py b/src/transformers/models/rt_detr/image_processing_rt_detr_fast.py
index 9f63b5b7ced4..0470352d38f4 100644
--- a/src/transformers/models/rt_detr/image_processing_rt_detr_fast.py
+++ b/src/transformers/models/rt_detr/image_processing_rt_detr_fast.py
@@ -43,7 +43,6 @@
     get_image_type,
     infer_channel_dimension_format,
     make_list_of_images,
-    pil_torch_interpolation_mapping,
     validate_annotations,
 )
 from ...utils import (
@@ -197,7 +196,7 @@ def __init__(
         format: Union[str, AnnotationFormat] = AnnotationFormat.COCO_DETECTION,
         do_resize: bool = True,
         size: Dict[str, int] = None,
-        resample: Union[PILImageResampling, F.InterpolationMode] = PILImageResampling.BILINEAR,
+        resample: Union[PILImageResampling, "F.InterpolationMode"] = PILImageResampling.BILINEAR,
         do_rescale: bool = True,
         rescale_factor: Union[int, float] = 1 / 255,
         do_normalize: bool = False,
@@ -256,7 +255,7 @@ def resize(
         self,
         image: torch.Tensor,
         size: SizeDict,
-        interpolation: F.InterpolationMode = F.InterpolationMode.BILINEAR,
+        interpolation: "F.InterpolationMode" = None,
         **kwargs,
     ) -> torch.Tensor:
         """
@@ -279,6 +278,7 @@ def resize(
             interpolation (`InterpolationMode`, *optional*, defaults to `InterpolationMode.BILINEAR`):
                 Resampling filter to use if resizing the image.
         """
+        interpolation = interpolation if interpolation is not None else F.InterpolationMode.BILINEAR
         if size.shortest_edge and size.longest_edge:
             # Resize the image so that the shortest edge or the longest edge is of the given size
             # while maintaining the aspect ratio of the original image.
@@ -312,7 +312,7 @@ def resize_annotation(
         orig_size: Tuple[int, int],
         target_size: Tuple[int, int],
         threshold: float = 0.5,
-        interpolation: F.InterpolationMode = F.InterpolationMode.NEAREST,
+        interpolation: "F.InterpolationMode" = None,
     ):
         """
         Resizes an annotation to a target size.
@@ -329,6 +329,7 @@ def resize_annotation(
             resample (`InterpolationMode`, defaults to `InterpolationMode.NEAREST`):
                 The resampling filter to use when resizing the masks.
         """
+        interpolation = interpolation if interpolation is not None else F.InterpolationMode.NEAREST
         ratio_height, ratio_width = [target / orig for target, orig in zip(target_size, orig_size)]
 
         new_annotation = {}
@@ -480,7 +481,7 @@ def preprocess(
         masks_path: Optional[Union[str, pathlib.Path]] = None,
         do_resize: Optional[bool] = None,
         size: Optional[Dict[str, int]] = None,
-        resample: Optional[Union[PILImageResampling, F.InterpolationMode]] = None,
+        resample: Optional[Union[PILImageResampling, "F.InterpolationMode"]] = None,
         do_rescale: Optional[bool] = None,
         rescale_factor: Optional[Union[int, float]] = None,
         do_normalize: Optional[bool] = None,