# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # 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"]