add MobileNetV2 model (#17845)

* add model files etc for MobileNetV2

* rename files for MobileNetV1

* initial implementation of MobileNetV1

* fix conversion script

* cleanup

* write docs

* tweaks

* fix conversion script

* extract hidden states

* fix test cases

* make fixup

* fixup it all

* rename V1 to V2

* fix checkpoints

* fixup

* implement first block + weight conversion

* add remaining layers

* add output stride and dilation

* fixup

* add tests

* add deeplabv3+ head

* a bit of fixup

* finish deeplab conversion

* add link to doc

* fix issue with JIT trace

in_height and in_width would be Tensor objects during JIT trace, which caused Core ML conversion to fail on the remainder op. By making them ints, the result of the padding calculation becomes a constant value.

* cleanup

* fix order of models

* fix rebase error

* remove main from doc link

* add image processor

* remove old feature extractor

* fix converter + other issues

* fixup

* fix unit test

* add to onnx tests (but these appear broken now)

* add post_process_semantic_segmentation

* use google org

* remove unused imports

* move args

* replace weird assert

src/transformers/models/mobilenet_v2/feature_extraction_mobilenet_v2.py

+# coding=utf-8
+# Copyright 2022 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.
+"""Feature extractor class for MobileNetV2."""
+
+from ...utils import logging
+from .image_processing_mobilenet_v2 import MobileNetV2ImageProcessor
+
+
+logger = logging.get_logger(__name__)
+
+
+MobileNetV2FeatureExtractor = MobileNetV2ImageProcessor

src/transformers/models/mobilenet_v2/image_processing_mobilenet_v2.py

+# coding=utf-8
+# Copyright 2022 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.
+"""Image processor class for MobileNetV2."""
+
+from typing import Dict, List, Optional, Tuple, Union
+
+import numpy as np
+
+from transformers.utils import is_torch_available, is_torch_tensor
+from transformers.utils.generic import TensorType
+
+from ...image_processing_utils import BaseImageProcessor, BatchFeature, get_size_dict
+from ...image_transforms import (
+    center_crop,
+    get_resize_output_image_size,
+    normalize,
+    rescale,
+    resize,
+    to_channel_dimension_format,
+)
+from ...image_utils import (
+    IMAGENET_STANDARD_MEAN,
+    IMAGENET_STANDARD_STD,
+    ChannelDimension,
+    ImageInput,
+    PILImageResampling,
+    is_batched,
+    to_numpy_array,
+    valid_images,
+)
+from ...utils import logging
+
+
+if is_torch_available():
+    import torch
+
+
+logger = logging.get_logger(__name__)
+
+
+class MobileNetV2ImageProcessor(BaseImageProcessor):
+    r"""
+    Constructs a MobileNetV2 image processor.
+
+    Args:
+        do_resize (`bool`, *optional*, defaults to `True`):
+            Whether to resize the image's (height, width) dimensions to the specified `size`. Can be overridden by
+            `do_resize` in the `preprocess` method.
+        size (`Dict[str, int]` *optional*, defaults to `{"shortest_edge": 256}`):
+            Size of the image after resizing. The shortest edge of the image is resized to size["shortest_edge"], with
+            the longest edge resized to keep the input aspect ratio. Can be overridden by `size` in the `preprocess`
+            method.
+        resample (`PILImageResampling`, *optional*, defaults to `PILImageResampling.BILINEAR`):
+            Resampling filter to use if resizing the image. Can be overridden by the `resample` parameter in the
+            `preprocess` method.
+        do_center_crop (`bool`, *optional*, defaults to `True`):
+            Whether to center crop the image. If the input size is smaller than `crop_size` along any edge, the image
+            is padded with 0's and then center cropped. Can be overridden by the `do_center_crop` parameter in the
+            `preprocess` method.
+        crop_size (`Dict[str, int]`, *optional*, defaults to `{"height": 224, "width": 224}`):
+            Desired output size when applying center-cropping. Only has an effect if `do_center_crop` is set to `True`.
+            Can be overridden by the `crop_size` parameter in the `preprocess` method.
+        do_rescale (`bool`, *optional*, defaults to `True`):
+            Whether to rescale the image by the specified scale `rescale_factor`. Can be overridden by the `do_rescale`
+            parameter in the `preprocess` method.
+        rescale_factor (`int` or `float`, *optional*, defaults to `1/255`):
+            Scale factor to use if rescaling the image. Can be overridden by the `rescale_factor` parameter in the
+            `preprocess` method.
+        do_normalize:
+            Whether to normalize the image. Can be overridden by the `do_normalize` parameter in the `preprocess`
+            method.
+        image_mean (`float` or `List[float]`, *optional*, defaults to `IMAGENET_STANDARD_MEAN`):
+            Mean to use if normalizing the image. This is a float or list of floats the length of the number of
+            channels in the image. Can be overridden by the `image_mean` parameter in the `preprocess` method.
+        image_std (`float` or `List[float]`, *optional*, defaults to `IMAGENET_STANDARD_STD`):
+            Standard deviation to use if normalizing the image. This is a float or list of floats the length of the
+            number of channels in the image. Can be overridden by the `image_std` parameter in the `preprocess` method.
+    """
+
+    model_input_names = ["pixel_values"]
+
+    def __init__(
+        self,
+        do_resize: bool = True,
+        size: Optional[Dict[str, int]] = None,
+        resample: PILImageResampling = PILImageResampling.BILINEAR,
+        do_center_crop: bool = True,
+        crop_size: Dict[str, int] = None,
+        do_rescale: bool = True,
+        rescale_factor: Union[int, float] = 1 / 255,
+        do_normalize: bool = True,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        **kwargs
+    ) -> None:
+        super().__init__(**kwargs)
+        size = size if size is not None else {"shortest_edge": 256}
+        size = get_size_dict(size, default_to_square=False)
+        crop_size = crop_size if crop_size is not None else {"height": 224, "width": 224}
+        crop_size = get_size_dict(crop_size)
+        self.do_resize = do_resize
+        self.size = size
+        self.resample = resample
+        self.do_center_crop = do_center_crop
+        self.crop_size = crop_size
+        self.do_rescale = do_rescale
+        self.rescale_factor = rescale_factor
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean if image_mean is not None else IMAGENET_STANDARD_MEAN
+        self.image_std = image_std if image_std is not None else IMAGENET_STANDARD_STD
+
+    def resize(
+        self,
+        image: np.ndarray,
+        size: Dict[str, int],
+        resample: PILImageResampling = PILImageResampling.BICUBIC,
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs
+    ) -> np.ndarray:
+        """
+        Resize an image. The shortest edge of the image is resized to size["shortest_edge"], with the longest edge
+        resized to keep the input aspect ratio.
+
+        Args:
+            image (`np.ndarray`):
+                Image to resize.
+            size (`Dict[str, int]`):
+                Size of the output image.
+            resample (`PILImageResampling`, *optional*, defaults to `PILImageResampling.BICUBIC`):
+                Resampling filter to use when resiizing the image.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format of the image. If not provided, it will be the same as the input image.
+        """
+        size = get_size_dict(size, default_to_square=False)
+        if "shortest_edge" not in size:
+            raise ValueError(f"The `size` parameter must contain the key `shortest_edge`. Got {size.keys()}")
+        output_size = get_resize_output_image_size(image, size=size["shortest_edge"], default_to_square=False)
+        return resize(image, size=output_size, resample=resample, data_format=data_format, **kwargs)
+
+    def center_crop(
+        self,
+        image: np.ndarray,
+        size: Dict[str, int],
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs
+    ) -> np.ndarray:
+        """
+        Center crop an image to (size["height"], size["width"]). If the input size is smaller than `size` along any
+        edge, the image is padded with 0's and then center cropped.
+
+        Args:
+            image (`np.ndarray`):
+                Image to center crop.
+            size (`Dict[str, int]`):
+                Size of the output image.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format of the image. If not provided, it will be the same as the input image.
+        """
+        size = get_size_dict(size)
+        return center_crop(image, size=(size["height"], size["width"]), data_format=data_format, **kwargs)
+
+    def rescale(
+        self, image: np.ndarray, scale: float, data_format: Optional[Union[str, ChannelDimension]] = None, **kwargs
+    ) -> np.ndarray:
+        """
+        Rescale an image by a scale factor. image = image * scale.
+
+        Args:
+            image (`np.ndarray`):
+                Image to rescale.
+            scale (`float`):
+                The scaling factor to rescale pixel values by.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format for the output image. If unset, the channel dimension format of the input
+                image is used. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+
+        Returns:
+            `np.ndarray`: The rescaled image.
+        """
+        return rescale(image, scale=scale, data_format=data_format, **kwargs)
+
+    def normalize(
+        self,
+        image: np.ndarray,
+        mean: Union[float, List[float]],
+        std: Union[float, List[float]],
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs
+    ) -> np.ndarray:
+        """
+        Normalize an image. image = (image - image_mean) / image_std.
+
+        Args:
+            image (`np.ndarray`):
+                Image to normalize.
+            mean (`float` or `List[float]`):
+                Image mean to use for normalization.
+            std (`float` or `List[float]`):
+                Image standard deviation to use for normalization.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format for the output image. If unset, the channel dimension format of the input
+                image is used. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+
+        Returns:
+            `np.ndarray`: The normalized image.
+        """
+        return normalize(image, mean=mean, std=std, data_format=data_format, **kwargs)
+
+    def preprocess(
+        self,
+        images: ImageInput,
+        do_resize: Optional[bool] = None,
+        size: Dict[str, int] = None,
+        resample: PILImageResampling = None,
+        do_center_crop: bool = None,
+        crop_size: Dict[str, int] = None,
+        do_rescale: Optional[bool] = None,
+        rescale_factor: Optional[float] = None,
+        do_normalize: Optional[bool] = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        data_format: Union[str, ChannelDimension] = ChannelDimension.FIRST,
+        **kwargs,
+    ):
+        """
+        Preprocess an image or batch of images.
+
+        Args:
+            images (`ImageInput`):
+                Image to preprocess.
+            do_resize (`bool`, *optional*, defaults to `self.do_resize`):
+                Whether to resize the image.
+            size (`Dict[str, int]`, *optional*, defaults to `self.size`):
+                Size of the image after resizing. Shortest edge of the image is resized to size["shortest_edge"], with
+                the longest edge resized to keep the input aspect ratio.
+            resample (`PILImageResampling` filter, *optional*, defaults to `self.resample`):
+                `PILImageResampling` filter to use if resizing the image e.g. `PILImageResampling.BILINEAR`. Only has
+                an effect if `do_resize` is set to `True`.
+            do_center_crop (`bool`, *optional*, defaults to `self.do_center_crop`):
+                Whether to center crop the image.
+            crop_size (`Dict[str, int]`, *optional*, defaults to `self.crop_size`):
+                Size of the center crop. Only has an effect if `do_center_crop` is set to `True`.
+            do_rescale (`bool`, *optional*, defaults to `self.do_rescale`):
+                Whether to rescale the image values between [0 - 1].
+            rescale_factor (`float`, *optional*, defaults to `self.rescale_factor`):
+                Rescale factor to rescale the image by if `do_rescale` is set to `True`.
+            do_normalize (`bool`, *optional*, defaults to `self.do_normalize`):
+                Whether to normalize the image.
+            image_mean (`float` or `List[float]`, *optional*, defaults to `self.image_mean`):
+                Image mean to use if `do_normalize` is set to `True`.
+            image_std (`float` or `List[float]`, *optional*, defaults to `self.image_std`):
+                Image standard deviation to use if `do_normalize` is set to `True`.
+            return_tensors (`str` or `TensorType`, *optional*):
+                The type of tensors to return. Can be one of:
+                - Unset: Return a list of `np.ndarray`.
+                - `TensorType.TENSORFLOW` or `'tf'`: Return a batch of type `tf.Tensor`.
+                - `TensorType.PYTORCH` or `'pt'`: Return a batch of type `torch.Tensor`.
+                - `TensorType.NUMPY` or `'np'`: Return a batch of type `np.ndarray`.
+                - `TensorType.JAX` or `'jax'`: Return a batch of type `jax.numpy.ndarray`.
+            data_format (`ChannelDimension` or `str`, *optional*, defaults to `ChannelDimension.FIRST`):
+                The channel dimension format for the output image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - Unset: Use the channel dimension format of the input image.
+        """
+        do_resize = do_resize if do_resize is not None else self.do_resize
+        size = size if size is not None else self.size
+        size = get_size_dict(size, default_to_square=False)
+        resample = resample if resample is not None else self.resample
+        do_center_crop = do_center_crop if do_center_crop is not None else self.do_center_crop
+        crop_size = crop_size if crop_size is not None else self.crop_size
+        crop_size = get_size_dict(crop_size)
+        do_rescale = do_rescale if do_rescale is not None else self.do_rescale
+        rescale_factor = rescale_factor if rescale_factor is not None else self.rescale_factor
+        do_normalize = do_normalize if do_normalize is not None else self.do_normalize
+        image_mean = image_mean if image_mean is not None else self.image_mean
+        image_std = image_std if image_std is not None else self.image_std
+
+        if not is_batched(images):
+            images = [images]
+
+        if not valid_images(images):
+            raise ValueError(
+                "Invalid image type. Must be of type PIL.Image.Image, numpy.ndarray, "
+                "torch.Tensor, tf.Tensor or jax.ndarray."
+            )
+
+        if do_resize and size is None:
+            raise ValueError("Size must be specified if do_resize is True.")
+
+        if do_center_crop and crop_size is None:
+            raise ValueError("Crop size must be specified if do_center_crop is True.")
+
+        if do_rescale and rescale_factor is None:
+            raise ValueError("Rescale factor must be specified if do_rescale is True.")
+
+        if do_normalize and (image_mean is None or image_std is None):
+            raise ValueError("Image mean and std must be specified if do_normalize is True.")
+
+        # All transformations expect numpy arrays.
+        images = [to_numpy_array(image) for image in images]
+
+        if do_resize:
+            images = [self.resize(image=image, size=size, resample=resample) for image in images]
+
+        if do_center_crop:
+            images = [self.center_crop(image=image, size=crop_size) for image in images]
+
+        if do_rescale:
+            images = [self.rescale(image=image, scale=rescale_factor) for image in images]
+
+        if do_normalize:
+            images = [self.normalize(image=image, mean=image_mean, std=image_std) for image in images]
+
+        images = [to_channel_dimension_format(image, data_format) for image in images]
+
+        data = {"pixel_values": images}
+        return BatchFeature(data=data, tensor_type=return_tensors)
+
+    def post_process_semantic_segmentation(self, outputs, target_sizes: List[Tuple] = None):
+        """
+        Converts the output of [`MobileNetV2ForSemanticSegmentation`] into semantic segmentation maps. Only supports
+        PyTorch.
+
+        Args:
+            outputs ([`MobileNetV2ForSemanticSegmentation`]):
+                Raw outputs of the model.
+            target_sizes (`List[Tuple]`, *optional*):
+                A list of length `batch_size`, where each item is a `Tuple[int, int]` corresponding to the requested
+                final size (height, width) of each prediction. If left to None, predictions will not be resized.
+        Returns:
+            `List[torch.Tensor]`:
+                A list 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

src/transformers/onnx/features.py

@@ -408,6 +408,11 @@ class FeaturesManager:
             "question-answering",
             onnx_config_cls="models.mobilebert.MobileBertOnnxConfig",
         ),
+        "mobilenet_v2": supported_features_mapping(
+            "default",
+            "image-classification",
+            onnx_config_cls="models.mobilenet_v2.MobileNetV2OnnxConfig",
+        ),
         "mobilevit": supported_features_mapping(
             "default",
             "image-classification",

src/transformers/utils/dummy_pt_objects.py

@@ -3526,6 +3526,41 @@ def load_tf_weights_in_mobilebert(*args, **kwargs):
     requires_backends(load_tf_weights_in_mobilebert, ["torch"])
 
 
+MOBILENET_V2_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class MobileNetV2ForImageClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileNetV2ForSemanticSegmentation(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileNetV2Model(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileNetV2PreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+def load_tf_weights_in_mobilenet_v2(*args, **kwargs):
+    requires_backends(load_tf_weights_in_mobilenet_v2, ["torch"])
+
+
 MOBILEVIT_PRETRAINED_MODEL_ARCHIVE_LIST = None
 
 

src/transformers/utils/dummy_vision_objects.py

@@ -225,6 +225,20 @@ class MaskFormerFeatureExtractor(metaclass=DummyObject):
         requires_backends(self, ["vision"])
 
 
+class MobileNetV2FeatureExtractor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
+class MobileNetV2ImageProcessor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
 class MobileViTFeatureExtractor(metaclass=DummyObject):
     _backends = ["vision"]
 

tests/models/mobilenet_v2/test_feature_extraction_mobilenet_v2.py

+# coding=utf-8
+# Copyright 2022 HuggingFace Inc.
+#
+# 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.
+
+
+import unittest
+
+import numpy as np
+
+from transformers.testing_utils import require_torch, require_vision
+from transformers.utils import is_torch_available, is_vision_available
+
+from ...test_feature_extraction_common import FeatureExtractionSavingTestMixin, prepare_image_inputs
+
+
+if is_torch_available():
+    import torch
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import MobileNetV2FeatureExtractor
+
+
+class MobileNetV2FeatureExtractionTester(unittest.TestCase):
+    def __init__(
+        self,
+        parent,
+        batch_size=7,
+        num_channels=3,
+        image_size=18,
+        min_resolution=30,
+        max_resolution=400,
+        do_resize=True,
+        size=None,
+        do_center_crop=True,
+        crop_size=None,
+    ):
+        size = size if size is not None else {"shortest_edge": 20}
+        crop_size = crop_size if crop_size is not None else {"height": 18, "width": 18}
+        self.parent = parent
+        self.batch_size = batch_size
+        self.num_channels = num_channels
+        self.image_size = image_size
+        self.min_resolution = min_resolution
+        self.max_resolution = max_resolution
+        self.do_resize = do_resize
+        self.size = size
+        self.do_center_crop = do_center_crop
+        self.crop_size = crop_size
+
+    def prepare_feat_extract_dict(self):
+        return {
+            "do_resize": self.do_resize,
+            "size": self.size,
+            "do_center_crop": self.do_center_crop,
+            "crop_size": self.crop_size,
+        }
+
+
+@require_torch
+@require_vision
+class MobileNetV2FeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCase):
+
+    feature_extraction_class = MobileNetV2FeatureExtractor if is_vision_available() else None
+
+    def setUp(self):
+        self.feature_extract_tester = MobileNetV2FeatureExtractionTester(self)
+
+    @property
+    def feat_extract_dict(self):
+        return self.feature_extract_tester.prepare_feat_extract_dict()
+
+    def test_feat_extract_properties(self):
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        self.assertTrue(hasattr(feature_extractor, "do_resize"))
+        self.assertTrue(hasattr(feature_extractor, "size"))
+        self.assertTrue(hasattr(feature_extractor, "do_center_crop"))
+        self.assertTrue(hasattr(feature_extractor, "center_crop"))
+
+    def test_batch_feature(self):
+        pass
+
+    def test_call_pil(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random PIL images
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False)
+        for image in image_inputs:
+            self.assertIsInstance(image, Image.Image)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                1,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size["height"],
+                self.feature_extract_tester.crop_size["width"],
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size["height"],
+                self.feature_extract_tester.crop_size["width"],
+            ),
+        )
+
+    def test_call_numpy(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random numpy tensors
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, numpify=True)
+        for image in image_inputs:
+            self.assertIsInstance(image, np.ndarray)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                1,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size["height"],
+                self.feature_extract_tester.crop_size["width"],
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size["height"],
+                self.feature_extract_tester.crop_size["width"],
+            ),
+        )
+
+    def test_call_pytorch(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random PyTorch tensors
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, torchify=True)
+        for image in image_inputs:
+            self.assertIsInstance(image, torch.Tensor)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                1,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size["height"],
+                self.feature_extract_tester.crop_size["width"],
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                self.feature_extract_tester.crop_size["height"],
+                self.feature_extract_tester.crop_size["width"],
+            ),
+        )

tests/models/mobilenet_v2/test_modeling_mobilenet_v2.py

+# coding=utf-8
+# Copyright 2022 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.
+""" Testing suite for the PyTorch MobileNetV2 model. """
+
+
+import inspect
+import unittest
+
+from transformers import MobileNetV2Config
+from transformers.testing_utils import require_torch, require_vision, slow, torch_device
+from transformers.utils import cached_property, is_torch_available, is_vision_available
+
+from ...test_configuration_common import ConfigTester
+from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import MobileNetV2ForImageClassification, MobileNetV2ForSemanticSegmentation, MobileNetV2Model
+    from transformers.models.mobilenet_v2.modeling_mobilenet_v2 import MOBILENET_V2_PRETRAINED_MODEL_ARCHIVE_LIST
+
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import MobileNetV2FeatureExtractor
+
+
+class MobileNetV2ConfigTester(ConfigTester):
+    def create_and_test_config_common_properties(self):
+        config = self.config_class(**self.inputs_dict)
+        self.parent.assertTrue(hasattr(config, "tf_padding"))
+        self.parent.assertTrue(hasattr(config, "depth_multiplier"))
+
+
+class MobileNetV2ModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        num_channels=3,
+        image_size=32,
+        depth_multiplier=0.25,
+        depth_divisible_by=8,
+        min_depth=8,
+        expand_ratio=6,
+        output_stride=32,
+        first_layer_is_expansion=True,
+        finegrained_output=True,
+        tf_padding=True,
+        hidden_act="relu6",
+        last_hidden_size=1280,
+        classifier_dropout_prob=0.1,
+        initializer_range=0.02,
+        is_training=True,
+        use_labels=True,
+        num_labels=10,
+        scope=None,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.num_channels = num_channels
+        self.image_size = image_size
+        self.depth_multiplier = depth_multiplier
+        self.depth_divisible_by = depth_divisible_by
+        self.min_depth = min_depth
+        self.expand_ratio = expand_ratio
+        self.tf_padding = tf_padding
+        self.output_stride = output_stride
+        self.first_layer_is_expansion = first_layer_is_expansion
+        self.finegrained_output = finegrained_output
+        self.hidden_act = hidden_act
+        self.last_hidden_size = last_hidden_size if finegrained_output else int(last_hidden_size * depth_multiplier)
+        self.classifier_dropout_prob = classifier_dropout_prob
+        self.use_labels = use_labels
+        self.is_training = is_training
+        self.num_labels = num_labels
+        self.initializer_range = initializer_range
+        self.scope = scope
+
+    def prepare_config_and_inputs(self):
+        pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
+
+        labels = None
+        pixel_labels = None
+        if self.use_labels:
+            labels = ids_tensor([self.batch_size], self.num_labels)
+            pixel_labels = ids_tensor([self.batch_size, self.image_size, self.image_size], self.num_labels)
+
+        config = self.get_config()
+
+        return config, pixel_values, labels, pixel_labels
+
+    def get_config(self):
+        return MobileNetV2Config(
+            num_channels=self.num_channels,
+            image_size=self.image_size,
+            depth_multiplier=self.depth_multiplier,
+            depth_divisible_by=self.depth_divisible_by,
+            min_depth=self.min_depth,
+            expand_ratio=self.expand_ratio,
+            output_stride=self.output_stride,
+            first_layer_is_expansion=self.first_layer_is_expansion,
+            finegrained_output=self.finegrained_output,
+            hidden_act=self.hidden_act,
+            tf_padding=self.tf_padding,
+            classifier_dropout_prob=self.classifier_dropout_prob,
+            initializer_range=self.initializer_range,
+        )
+
+    def create_and_check_model(self, config, pixel_values, labels, pixel_labels):
+        model = MobileNetV2Model(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(pixel_values)
+        self.parent.assertEqual(
+            result.last_hidden_state.shape,
+            (
+                self.batch_size,
+                self.last_hidden_size,
+                self.image_size // self.output_stride,
+                self.image_size // self.output_stride,
+            ),
+        )
+        self.parent.assertEqual(
+            result.pooler_output.shape,
+            (self.batch_size, self.last_hidden_size),
+        )
+
+    def create_and_check_for_image_classification(self, config, pixel_values, labels, pixel_labels):
+        config.num_labels = self.num_labels
+        model = MobileNetV2ForImageClassification(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(pixel_values, labels=labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
+
+    def create_and_check_for_semantic_segmentation(self, config, pixel_values, labels, pixel_labels):
+        config.num_labels = self.num_labels
+        model = MobileNetV2ForSemanticSegmentation(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(pixel_values)
+        self.parent.assertEqual(
+            result.logits.shape,
+            (
+                self.batch_size,
+                self.num_labels,
+                self.image_size // self.output_stride,
+                self.image_size // self.output_stride,
+            ),
+        )
+        result = model(pixel_values, labels=pixel_labels)
+        self.parent.assertEqual(
+            result.logits.shape,
+            (
+                self.batch_size,
+                self.num_labels,
+                self.image_size // self.output_stride,
+                self.image_size // self.output_stride,
+            ),
+        )
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        config, pixel_values, labels, pixel_labels = config_and_inputs
+        inputs_dict = {"pixel_values": pixel_values}
+        return config, inputs_dict
+
+
+@require_torch
+class MobileNetV2ModelTest(ModelTesterMixin, unittest.TestCase):
+    """
+    Here we also overwrite some of the tests of test_modeling_common.py, as MobileNetV2 does not use input_ids, inputs_embeds,
+    attention_mask and seq_length.
+    """
+
+    all_model_classes = (
+        (MobileNetV2Model, MobileNetV2ForImageClassification, MobileNetV2ForSemanticSegmentation)
+        if is_torch_available()
+        else ()
+    )
+
+    test_pruning = False
+    test_resize_embeddings = False
+    test_head_masking = False
+    has_attentions = False
+
+    def setUp(self):
+        self.model_tester = MobileNetV2ModelTester(self)
+        self.config_tester = MobileNetV2ConfigTester(self, config_class=MobileNetV2Config, has_text_modality=False)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    @unittest.skip(reason="MobileNetV2 does not use inputs_embeds")
+    def test_inputs_embeds(self):
+        pass
+
+    @unittest.skip(reason="MobileNetV2 does not support input and output embeddings")
+    def test_model_common_attributes(self):
+        pass
+
+    @unittest.skip(reason="MobileNetV2 does not output attentions")
+    def test_attention_outputs(self):
+        pass
+
+    def test_forward_signature(self):
+        config, _ = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            model = model_class(config)
+            signature = inspect.signature(model.forward)
+            # signature.parameters is an OrderedDict => so arg_names order is deterministic
+            arg_names = [*signature.parameters.keys()]
+
+            expected_arg_names = ["pixel_values"]
+            self.assertListEqual(arg_names[:1], expected_arg_names)
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_hidden_states_output(self):
+        def check_hidden_states_output(inputs_dict, config, model_class):
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            hidden_states = outputs.hidden_states
+
+            expected_num_stages = 16
+            self.assertEqual(len(hidden_states), expected_num_stages)
+
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            inputs_dict["output_hidden_states"] = True
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+            # check that output_hidden_states also work using config
+            del inputs_dict["output_hidden_states"]
+            config.output_hidden_states = True
+
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+    def test_for_image_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_image_classification(*config_and_inputs)
+
+    def test_for_semantic_segmentation(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_semantic_segmentation(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in MOBILENET_V2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = MobileNetV2Model.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+
+# We will verify our results on an image of cute cats
+def prepare_img():
+    image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+    return image
+
+
+@require_torch
+@require_vision
+class MobileNetV2ModelIntegrationTest(unittest.TestCase):
+    @cached_property
+    def default_feature_extractor(self):
+        return (
+            MobileNetV2FeatureExtractor.from_pretrained("google/mobilenet_v2_1.0_224")
+            if is_vision_available()
+            else None
+        )
+
+    @slow
+    def test_inference_image_classification_head(self):
+        model = MobileNetV2ForImageClassification.from_pretrained("google/mobilenet_v2_1.0_224").to(torch_device)
+
+        feature_extractor = self.default_feature_extractor
+        image = prepare_img()
+        inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device)
+
+        # forward pass
+        with torch.no_grad():
+            outputs = model(**inputs)
+
+        # verify the logits
+        expected_shape = torch.Size((1, 1001))
+        self.assertEqual(outputs.logits.shape, expected_shape)
+
+        expected_slice = torch.tensor([0.2445, -1.1993, 0.1905]).to(torch_device)
+
+        self.assertTrue(torch.allclose(outputs.logits[0, :3], expected_slice, atol=1e-4))
+
+    @slow
+    def test_inference_semantic_segmentation(self):
+        model = MobileNetV2ForSemanticSegmentation.from_pretrained("google/deeplabv3_mobilenet_v2_1.0_513")
+        model = model.to(torch_device)
+
+        feature_extractor = MobileNetV2FeatureExtractor.from_pretrained("google/deeplabv3_mobilenet_v2_1.0_513")
+
+        image = prepare_img()
+        inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device)
+
+        # forward pass
+        with torch.no_grad():
+            outputs = model(**inputs)
+        logits = outputs.logits
+
+        # verify the logits
+        expected_shape = torch.Size((1, 21, 65, 65))
+        self.assertEqual(logits.shape, expected_shape)
+
+        expected_slice = torch.tensor(
+            [
+                [[17.5790, 17.7581, 18.3355], [18.3257, 18.4230, 18.8973], [18.6169, 18.8650, 19.2187]],
+                [[-2.1595, -2.0977, -2.3741], [-2.4226, -2.3028, -2.6835], [-2.7819, -2.5991, -2.7706]],
+                [[4.2058, 4.8317, 4.7638], [4.4136, 5.0361, 4.9383], [4.5028, 4.9644, 4.8734]],
+            ],
+            device=torch_device,
+        )
+
+        self.assertTrue(torch.allclose(logits[0, :3, :3, :3], expected_slice, atol=1e-4))

tests/onnx/test_onnx_v2.py

@@ -199,6 +199,7 @@ PYTORCH_EXPORT_MODELS = {
     ("roformer", "junnyu/roformer_chinese_base"),
     ("squeezebert", "squeezebert/squeezebert-uncased"),
     ("mobilebert", "google/mobilebert-uncased"),
+    ("mobilenet_v2", "google/mobilenet_v2_0.35_96"),
     ("mobilevit", "apple/mobilevit-small"),
     ("xlm", "xlm-clm-ende-1024"),
     ("xlm-roberta", "xlm-roberta-base"),

utils/documentation_tests.txt

@@ -107,6 +107,7 @@ src/transformers/models/megatron_bert/configuration_megatron_bert.py
 src/transformers/models/mobilebert/configuration_mobilebert.py
 src/transformers/models/mobilebert/modeling_mobilebert.py
 src/transformers/models/mobilebert/modeling_tf_mobilebert.py
+src/transformers/models/mobilenet_v2/modeling_mobilenet_v2.py
 src/transformers/models/mobilevit/modeling_mobilevit.py
 src/transformers/models/mobilevit/modeling_tf_mobilevit.py
 src/transformers/models/nezha/configuration_nezha.py