Add Fast Segformer Processor (#37024)

* Add Fast Segformer Processor

* Modified the params according to segformer model

* modified test_image_processing_Segformer_fast args

- removed redundant params like do_center_crop,center_crop which aren't present in the original segformer class

* added segmentation_maps processing logic form the slow segformer processing module with references from beitimageprocessing fast

* fixed code_quality

* added recommended fixes and tests to make sure everything processess smoothly

* Fixed SegmentationMapsLogic

- modified the preprocessing of segmentation maps to use tensors
- added batch support

* fixed some mismatched files

* modified the tolerance for tests

* use modular

* fix ci

---------

Co-authored-by: yonigozlan <yoni.gozlan@huggingface.co>

docs/source/en/model_doc/segformer.md

@@ -128,6 +128,12 @@ If you're interested in submitting a resource to be included here, please feel f
     - preprocess
     - post_process_semantic_segmentation
 
+## SegformerImageProcessorFast
+
+[[autodoc]] SegformerImageProcessorFast
+    - preprocess
+    - post_process_semantic_segmentation
+
 <frameworkcontent>
 <pt>
 

src/transformers/models/auto/image_processing_auto.py

@@ -156,6 +156,7 @@ else:
             ("sam", ("SamImageProcessor", "SamImageProcessorFast")),
             ("sam_hq", ("SamImageProcessor", "SamImageProcessorFast")),
             ("segformer", ("SegformerImageProcessor",)),
+            ("segformer", ("SegformerImageProcessor", "SegformerImageProcessorFast")),
             ("seggpt", ("SegGptImageProcessor",)),
             ("shieldgemma2", ("Gemma3ImageProcessor", "Gemma3ImageProcessorFast")),
             ("siglip", ("SiglipImageProcessor", "SiglipImageProcessorFast")),
@@ -179,7 +180,8 @@ else:
             ("tvlt", ("TvltImageProcessor",)),
             ("tvp", ("TvpImageProcessor",)),
             ("udop", ("LayoutLMv3ImageProcessor", "LayoutLMv3ImageProcessorFast")),
-            ("upernet", ("SegformerImageProcessor",)),
+            ("udop", ("LayoutLMv3ImageProcessor",)),
+            ("upernet", ("SegformerImageProcessor", "SegformerImageProcessorFast")),
             ("van", ("ConvNextImageProcessor", "ConvNextImageProcessorFast")),
             ("videomae", ("VideoMAEImageProcessor",)),
             ("vilt", ("ViltImageProcessor", "ViltImageProcessorFast")),

src/transformers/models/beit/image_processing_beit_fast.py

@@ -16,9 +16,6 @@
 
 from typing import Optional, Union
 
-import torch
-from torchvision.transforms import functional as F
-
 from ...image_processing_utils import BatchFeature
 from ...image_processing_utils_fast import (
     BaseImageProcessorFast,
@@ -36,11 +33,26 @@ from ...image_utils import (
     is_torch_tensor,
 )
 from ...processing_utils import Unpack
-from ...utils import TensorType, auto_docstring
+from ...utils import (
+    TensorType,
+    auto_docstring,
+    is_torch_available,
+    is_torchvision_available,
+    is_torchvision_v2_available,
+)
+
+
+if is_torch_available():
+    import torch
+
+if is_torchvision_v2_available():
+    from torchvision.transforms.v2 import functional as F
+elif is_torchvision_available():
+    from torchvision.transforms import functional as F
 
 
 class BeitFastImageProcessorKwargs(DefaultFastImageProcessorKwargs):
-    """
+    r"""
     do_reduce_labels (`bool`, *optional*, defaults to `self.do_reduce_labels`):
         Whether or not to reduce all label values of segmentation maps by 1. Usually used for datasets where 0
         is used for background, and background itself is not included in all classes of a dataset (e.g.

src/transformers/models/segformer/__init__.py

@@ -21,6 +21,7 @@ if TYPE_CHECKING:
     from .configuration_segformer import *
     from .feature_extraction_segformer import *
     from .image_processing_segformer import *
+    from .image_processing_segformer_fast import *
     from .modeling_segformer import *
     from .modeling_tf_segformer import *
 else:

src/transformers/models/segformer/image_processing_segformer_fast.py

@@ -0,0 +1,247 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/segformer/modular_segformer.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_segformer.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional, Union
+
+from ...image_processing_utils import BatchFeature
+from ...image_processing_utils_fast import (
+    BaseImageProcessorFast,
+    DefaultFastImageProcessorKwargs,
+    group_images_by_shape,
+    reorder_images,
+)
+from ...image_utils import (
+    IMAGENET_DEFAULT_MEAN,
+    IMAGENET_DEFAULT_STD,
+    ChannelDimension,
+    ImageInput,
+    PILImageResampling,
+    SizeDict,
+    is_torch_tensor,
+    pil_torch_interpolation_mapping,
+)
+from ...processing_utils import Unpack
+from ...utils import (
+    TensorType,
+    auto_docstring,
+    is_torch_available,
+    is_torchvision_available,
+    is_torchvision_v2_available,
+)
+
+
+if is_torch_available():
+    import torch
+
+if is_torchvision_v2_available():
+    from torchvision.transforms.v2 import functional as F
+elif is_torchvision_available():
+    from torchvision.transforms import functional as F
+
+
+class SegformerFastImageProcessorKwargs(DefaultFastImageProcessorKwargs):
+    r"""
+    do_reduce_labels (`bool`, *optional*, defaults to `self.do_reduce_labels`):
+        Whether or not to reduce all label values of segmentation maps by 1. Usually used for datasets where 0
+        is used for background, and background itself is not included in all classes of a dataset (e.g.
+        ADE20k). The background label will be replaced by 255.
+    """
+
+    do_reduce_labels: Optional[bool]
+
+
+@auto_docstring
+class SegformerImageProcessorFast(BaseImageProcessorFast):
+    resample = PILImageResampling.BILINEAR
+    image_mean = IMAGENET_DEFAULT_MEAN
+    image_std = IMAGENET_DEFAULT_STD
+    size = {"height": 512, "width": 512}
+    default_to_square = True
+    crop_size = None
+    do_resize = True
+    do_center_crop = None
+    do_rescale = True
+    do_normalize = True
+    do_reduce_labels = False
+    valid_kwargs = SegformerFastImageProcessorKwargs
+    rescale_factor = 1 / 255
+
+    def __init__(self, **kwargs: Unpack[SegformerFastImageProcessorKwargs]):
+        super().__init__(**kwargs)
+
+    def reduce_label(self, labels: list["torch.Tensor"]):
+        for idx in range(len(labels)):
+            label = labels[idx]
+            label = torch.where(label == 0, torch.tensor(255, dtype=label.dtype), label)
+            label = label - 1
+            label = torch.where(label == 254, torch.tensor(255, dtype=label.dtype), label)
+            labels[idx] = label
+
+        return label
+
+    @auto_docstring
+    def preprocess(
+        self,
+        images: ImageInput,
+        segmentation_maps: Optional[ImageInput] = None,
+        **kwargs: Unpack[SegformerFastImageProcessorKwargs],
+    ) -> BatchFeature:
+        r"""
+        segmentation_maps (`ImageInput`, *optional*):
+            The segmentation maps to preprocess.
+        """
+        return super().preprocess(images, segmentation_maps, **kwargs)
+
+    def _preprocess_image_like_inputs(
+        self,
+        images: ImageInput,
+        segmentation_maps: Optional[ImageInput],
+        do_convert_rgb: bool,
+        input_data_format: ChannelDimension,
+        device: Optional[Union[str, "torch.device"]] = None,
+        **kwargs: Unpack[SegformerFastImageProcessorKwargs],
+    ) -> BatchFeature:
+        """
+        Preprocess image-like inputs.
+        """
+        images = self._prepare_image_like_inputs(
+            images=images, do_convert_rgb=do_convert_rgb, input_data_format=input_data_format, device=device
+        )
+        images_kwargs = kwargs.copy()
+        images_kwargs["do_reduce_labels"] = False
+        batch_feature = self._preprocess(images, **images_kwargs)
+
+        if segmentation_maps is not None:
+            processed_segmentation_maps = self._prepare_image_like_inputs(
+                images=segmentation_maps,
+                expected_ndims=2,
+                do_convert_rgb=False,
+                input_data_format=ChannelDimension.FIRST,
+            )
+
+            segmentation_maps_kwargs = kwargs.copy()
+            segmentation_maps_kwargs.update(
+                {
+                    "do_normalize": False,
+                    "do_rescale": False,
+                    # Nearest interpolation is used for segmentation maps instead of BILINEAR.
+                    "interpolation": pil_torch_interpolation_mapping[PILImageResampling.NEAREST],
+                }
+            )
+            processed_segmentation_maps = self._preprocess(
+                images=processed_segmentation_maps, **segmentation_maps_kwargs
+            ).pixel_values
+            batch_feature["labels"] = processed_segmentation_maps.squeeze(1).to(torch.int64)
+
+        return batch_feature
+
+    def _preprocess(
+        self,
+        images: list["torch.Tensor"],
+        do_reduce_labels: bool,
+        interpolation: Optional["F.InterpolationMode"],
+        do_resize: bool,
+        do_rescale: bool,
+        do_normalize: bool,
+        size: SizeDict,
+        rescale_factor: float,
+        image_mean: Union[float, list[float]],
+        image_std: Union[float, list[float]],
+        disable_grouping: bool,
+        return_tensors: Optional[Union[str, TensorType]],
+        **kwargs,
+    ) -> BatchFeature:  # Return type can be list if return_tensors=None
+        if do_reduce_labels:
+            images = self.reduce_label(images)  # Apply reduction if needed
+
+        # Group images by size for batched resizing
+        resized_images = images
+        if do_resize:
+            grouped_images, grouped_images_index = group_images_by_shape(images, disable_grouping=disable_grouping)
+            resized_images_grouped = {}
+            for shape, stacked_images in grouped_images.items():
+                resized_stacked_images = self.resize(image=stacked_images, size=size, interpolation=interpolation)
+                resized_images_grouped[shape] = resized_stacked_images
+            resized_images = reorder_images(resized_images_grouped, grouped_images_index)
+
+        # Group images by size for further processing (rescale/normalize)
+        # Needed in case do_resize is False, or resize returns images with different sizes
+        grouped_images, grouped_images_index = group_images_by_shape(resized_images, disable_grouping=disable_grouping)
+        processed_images_grouped = {}
+        for shape, stacked_images in grouped_images.items():
+            # Fused rescale and normalize
+            stacked_images = self.rescale_and_normalize(
+                stacked_images, do_rescale, rescale_factor, do_normalize, image_mean, image_std
+            )
+            processed_images_grouped[shape] = stacked_images
+
+        processed_images = reorder_images(processed_images_grouped, grouped_images_index)
+
+        # Stack images into a single tensor if return_tensors is set
+        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
+
+        return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
+
+    def post_process_semantic_segmentation(self, outputs, target_sizes: Optional[list[tuple]] = None):
+        """
+        Converts the output of [`SegformerForSemanticSegmentation`] into semantic segmentation maps. Only supports PyTorch.
+
+        Args:
+            outputs ([`SegformerForSemanticSegmentation`]):
+                Raw outputs of the model.
+            target_sizes (`list[Tuple]` of length `batch_size`, *optional*):
+                List of tuples corresponding to the requested final size (height, width) of each prediction. If unset,
+                predictions will not be resized.
+
+        Returns:
+            semantic_segmentation: `list[torch.Tensor]` of length `batch_size`, where each item is a semantic
+            segmentation map of shape (height, width) corresponding to the target_sizes entry (if `target_sizes` is
+            specified). Each entry of each `torch.Tensor` correspond to a semantic class id.
+        """
+        # TODO: add support for other frameworks
+        logits = outputs.logits
+
+        # Resize logits and compute semantic segmentation maps
+        if target_sizes is not None:
+            if len(logits) != len(target_sizes):
+                raise ValueError(
+                    "Make sure that you pass in as many target sizes as the batch dimension of the logits"
+                )
+
+            if is_torch_tensor(target_sizes):
+                target_sizes = target_sizes.numpy()
+
+            semantic_segmentation = []
+
+            for idx in range(len(logits)):
+                resized_logits = torch.nn.functional.interpolate(
+                    logits[idx].unsqueeze(dim=0), size=target_sizes[idx], mode="bilinear", align_corners=False
+                )
+                semantic_map = resized_logits[0].argmax(dim=0)
+                semantic_segmentation.append(semantic_map)
+        else:
+            semantic_segmentation = logits.argmax(dim=1)
+            semantic_segmentation = [semantic_segmentation[i] for i in range(semantic_segmentation.shape[0])]
+
+        return semantic_segmentation
+
+
+__all__ = ["SegformerImageProcessorFast"]

src/transformers/models/segformer/modular_segformer.py

@@ -0,0 +1,161 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Fast Image processor class for Segformer."""
+
+from typing import Optional, Union
+
+from transformers.models.beit.image_processing_beit_fast import BeitFastImageProcessorKwargs, BeitImageProcessorFast
+
+from ...image_processing_utils import BatchFeature
+from ...image_processing_utils_fast import (
+    group_images_by_shape,
+    reorder_images,
+)
+from ...image_utils import (
+    IMAGENET_DEFAULT_MEAN,
+    IMAGENET_DEFAULT_STD,
+    ChannelDimension,
+    ImageInput,
+    PILImageResampling,
+    SizeDict,
+    pil_torch_interpolation_mapping,
+)
+from ...processing_utils import Unpack
+from ...utils import (
+    TensorType,
+    is_torch_available,
+    is_torchvision_available,
+    is_torchvision_v2_available,
+)
+
+
+if is_torch_available():
+    import torch
+
+if is_torchvision_v2_available():
+    from torchvision.transforms.v2 import functional as F
+elif is_torchvision_available():
+    from torchvision.transforms import functional as F
+
+
+class SegformerFastImageProcessorKwargs(BeitFastImageProcessorKwargs):
+    pass
+
+
+class SegformerImageProcessorFast(BeitImageProcessorFast):
+    resample = PILImageResampling.BILINEAR
+    image_mean = IMAGENET_DEFAULT_MEAN
+    image_std = IMAGENET_DEFAULT_STD
+    size = {"height": 512, "width": 512}
+    do_resize = True
+    do_rescale = True
+    rescale_factor = 1 / 255
+    do_normalize = True
+    do_reduce_labels = False
+    do_center_crop = None
+    crop_size = None
+
+    def _preprocess_image_like_inputs(
+        self,
+        images: ImageInput,
+        segmentation_maps: Optional[ImageInput],
+        do_convert_rgb: bool,
+        input_data_format: ChannelDimension,
+        device: Optional[Union[str, "torch.device"]] = None,
+        **kwargs: Unpack[SegformerFastImageProcessorKwargs],
+    ) -> BatchFeature:
+        """
+        Preprocess image-like inputs.
+        """
+        images = self._prepare_image_like_inputs(
+            images=images, do_convert_rgb=do_convert_rgb, input_data_format=input_data_format, device=device
+        )
+        images_kwargs = kwargs.copy()
+        images_kwargs["do_reduce_labels"] = False
+        batch_feature = self._preprocess(images, **images_kwargs)
+
+        if segmentation_maps is not None:
+            processed_segmentation_maps = self._prepare_image_like_inputs(
+                images=segmentation_maps,
+                expected_ndims=2,
+                do_convert_rgb=False,
+                input_data_format=ChannelDimension.FIRST,
+            )
+
+            segmentation_maps_kwargs = kwargs.copy()
+            segmentation_maps_kwargs.update(
+                {
+                    "do_normalize": False,
+                    "do_rescale": False,
+                    # Nearest interpolation is used for segmentation maps instead of BILINEAR.
+                    "interpolation": pil_torch_interpolation_mapping[PILImageResampling.NEAREST],
+                }
+            )
+            processed_segmentation_maps = self._preprocess(
+                images=processed_segmentation_maps, **segmentation_maps_kwargs
+            ).pixel_values
+            batch_feature["labels"] = processed_segmentation_maps.squeeze(1).to(torch.int64)
+
+        return batch_feature
+
+    def _preprocess(
+        self,
+        images: list["torch.Tensor"],
+        do_reduce_labels: bool,
+        interpolation: Optional["F.InterpolationMode"],
+        do_resize: bool,
+        do_rescale: bool,
+        do_normalize: bool,
+        size: SizeDict,
+        rescale_factor: float,
+        image_mean: Union[float, list[float]],
+        image_std: Union[float, list[float]],
+        disable_grouping: bool,
+        return_tensors: Optional[Union[str, TensorType]],
+        **kwargs,
+    ) -> BatchFeature:  # Return type can be list if return_tensors=None
+        if do_reduce_labels:
+            images = self.reduce_label(images)  # Apply reduction if needed
+
+        # Group images by size for batched resizing
+        resized_images = images
+        if do_resize:
+            grouped_images, grouped_images_index = group_images_by_shape(images, disable_grouping=disable_grouping)
+            resized_images_grouped = {}
+            for shape, stacked_images in grouped_images.items():
+                resized_stacked_images = self.resize(image=stacked_images, size=size, interpolation=interpolation)
+                resized_images_grouped[shape] = resized_stacked_images
+            resized_images = reorder_images(resized_images_grouped, grouped_images_index)
+
+        # Group images by size for further processing (rescale/normalize)
+        # Needed in case do_resize is False, or resize returns images with different sizes
+        grouped_images, grouped_images_index = group_images_by_shape(resized_images, disable_grouping=disable_grouping)
+        processed_images_grouped = {}
+        for shape, stacked_images in grouped_images.items():
+            # Fused rescale and normalize
+            stacked_images = self.rescale_and_normalize(
+                stacked_images, do_rescale, rescale_factor, do_normalize, image_mean, image_std
+            )
+            processed_images_grouped[shape] = stacked_images
+
+        processed_images = reorder_images(processed_images_grouped, grouped_images_index)
+
+        # Stack images into a single tensor if return_tensors is set
+        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
+
+        return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
+
+
+__all__ = ["SegformerImageProcessorFast"]

tests/models/segformer/test_image_processing_segformer.py

@@ -18,7 +18,7 @@ import unittest
 from datasets import load_dataset
 
 from transformers.testing_utils import require_torch, require_vision
-from transformers.utils import is_torch_available, is_vision_available
+from transformers.utils import is_torch_available, is_torchvision_available, is_vision_available
 
 from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
 
@@ -29,6 +29,9 @@ if is_torch_available():
 if is_vision_available():
     from transformers import SegformerImageProcessor
 
+    if is_torchvision_available():
+        from transformers import SegformerImageProcessorFast
+
 
 class SegformerImageProcessingTester:
     def __init__(
@@ -98,6 +101,7 @@ def prepare_semantic_batch_inputs():
 @require_vision
 class SegformerImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
     image_processing_class = SegformerImageProcessor if is_vision_available() else None
+    fast_image_processing_class = SegformerImageProcessorFast if is_torchvision_available() else None
 
     def setUp(self):
         super().setUp()
@@ -108,7 +112,8 @@ class SegformerImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
         return self.image_processor_tester.prepare_image_processor_dict()
 
     def test_image_processor_properties(self):
-        image_processing = self.image_processing_class(**self.image_processor_dict)
+        for image_processing_class in self.image_processor_list:
+            image_processing = image_processing_class(**self.image_processor_dict)
             self.assertTrue(hasattr(image_processing, "do_resize"))
             self.assertTrue(hasattr(image_processing, "size"))
             self.assertTrue(hasattr(image_processing, "do_normalize"))
@@ -117,19 +122,21 @@ class SegformerImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
             self.assertTrue(hasattr(image_processing, "do_reduce_labels"))
 
     def test_image_processor_from_dict_with_kwargs(self):
-        image_processor = self.image_processing_class.from_dict(self.image_processor_dict)
+        for image_processing_class in self.image_processor_list:
+            image_processor = image_processing_class.from_dict(self.image_processor_dict)
             self.assertEqual(image_processor.size, {"height": 30, "width": 30})
             self.assertEqual(image_processor.do_reduce_labels, False)
 
-        image_processor = self.image_processing_class.from_dict(
+            image_processor = image_processing_class.from_dict(
                 self.image_processor_dict, size=42, do_reduce_labels=True
             )
             self.assertEqual(image_processor.size, {"height": 42, "width": 42})
             self.assertEqual(image_processor.do_reduce_labels, True)
 
     def test_call_segmentation_maps(self):
+        for image_processing_class in self.image_processor_list:
             # Initialize image_processing
-        image_processing = self.image_processing_class(**self.image_processor_dict)
+            image_processing = image_processing_class(**self.image_processor_dict)
             # create random PyTorch tensors
             image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, torchify=True)
             maps = []
@@ -235,7 +242,8 @@ class SegformerImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
 
     def test_reduce_labels(self):
         # Initialize image_processing
-        image_processing = self.image_processing_class(**self.image_processor_dict)
+        for image_processing_class in self.image_processor_list:
+            image_processing = image_processing_class(**self.image_processor_dict)
 
             # ADE20k has 150 classes, and the background is included, so labels should be between 0 and 150
             image, map = prepare_semantic_single_inputs()
@@ -247,3 +255,48 @@ class SegformerImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
             encoding = image_processing(image, map, return_tensors="pt")
             self.assertTrue(encoding["labels"].min().item() >= 0)
             self.assertTrue(encoding["labels"].max().item() <= 255)
+
+    def test_slow_fast_equivalence(self):
+        if not self.test_slow_image_processor or not self.test_fast_image_processor:
+            self.skipTest(reason="Skipping slow/fast equivalence test")
+
+        if self.image_processing_class is None or self.fast_image_processing_class is None:
+            self.skipTest(reason="Skipping slow/fast equivalence test as one of the image processors is not defined")
+
+        dummy_image, dummy_map = prepare_semantic_single_inputs()
+
+        image_processor_slow = self.image_processing_class(**self.image_processor_dict)
+        image_processor_fast = self.fast_image_processing_class(**self.image_processor_dict)
+
+        image_encoding_slow = image_processor_slow(dummy_image, segmentation_maps=dummy_map, return_tensors="pt")
+        image_encoding_fast = image_processor_fast(dummy_image, segmentation_maps=dummy_map, return_tensors="pt")
+
+        self._assert_slow_fast_tensors_equivalence(image_encoding_slow.pixel_values, image_encoding_fast.pixel_values)
+        self._assert_slow_fast_tensors_equivalence(
+            image_encoding_slow.labels.float(), image_encoding_fast.labels.float(), atol=5, mean_atol=0.01
+        )
+
+    def test_slow_fast_equivalence_batched(self):
+        if not self.test_slow_image_processor or not self.test_fast_image_processor:
+            self.skipTest(reason="Skipping slow/fast equivalence test")
+
+        if self.image_processing_class is None or self.fast_image_processing_class is None:
+            self.skipTest(reason="Skipping slow/fast equivalence test as one of the image processors is not defined")
+
+        if hasattr(self.image_processor_tester, "do_center_crop") and self.image_processor_tester.do_center_crop:
+            self.skipTest(
+                reason="Skipping as do_center_crop is True and center_crop functions are not equivalent for fast and slow processors"
+            )
+
+        dummy_images, dummy_maps = prepare_semantic_batch_inputs()
+
+        image_processor_slow = self.image_processing_class(**self.image_processor_dict)
+        image_processor_fast = self.fast_image_processing_class(**self.image_processor_dict)
+
+        encoding_slow = image_processor_slow(dummy_images, segmentation_maps=dummy_maps, return_tensors="pt")
+        encoding_fast = image_processor_fast(dummy_images, segmentation_maps=dummy_maps, return_tensors="pt")
+
+        self._assert_slow_fast_tensors_equivalence(encoding_slow.pixel_values, encoding_fast.pixel_values)
+        self._assert_slow_fast_tensors_equivalence(
+            encoding_slow.labels.float(), encoding_fast.labels.float(), atol=5, mean_atol=0.01
+        )