add MobileNetV1 model (#17799)

* 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

remove main from doc link

fixes

fix tests

fix up

use google org

fix weird assert

* fixup

* use google organization for checkpoints

src/transformers/models/mobilenet_v1/image_processing_mobilenet_v1.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 MobileNetV1."""
+
+from typing import Dict, List, Optional, Union
+
+import numpy as np
+
+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
+
+
+logger = logging.get_logger(__name__)
+
+
+class MobileNetV1ImageProcessor(BaseImageProcessor):
+    r"""
+    Constructs a MobileNetV1 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)

src/transformers/models/mobilenet_v1/modeling_mobilenet_v1.py

+# coding=utf-8
+# Copyright 2022 Apple Inc. and 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.
+""" PyTorch MobileNetV1 model."""
+
+
+from typing import Optional, Union
+
+import torch
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from ...activations import ACT2FN
+from ...modeling_outputs import BaseModelOutputWithPoolingAndNoAttention, ImageClassifierOutputWithNoAttention
+from ...modeling_utils import PreTrainedModel
+from ...utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging
+from .configuration_mobilenet_v1 import MobileNetV1Config
+
+
+logger = logging.get_logger(__name__)
+
+
+# General docstring
+_CONFIG_FOR_DOC = "MobileNetV1Config"
+_FEAT_EXTRACTOR_FOR_DOC = "MobileNetV1FeatureExtractor"
+
+# Base docstring
+_CHECKPOINT_FOR_DOC = "google/mobilenet_v1_1.0_224"
+_EXPECTED_OUTPUT_SHAPE = [1, 1024, 7, 7]
+
+# Image classification docstring
+_IMAGE_CLASS_CHECKPOINT = "google/mobilenet_v1_1.0_224"
+_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
+
+
+MOBILENET_V1_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "google/mobilenet_v1_1.0_224",
+    "google/mobilenet_v1_0.75_192",
+    # See all MobileNetV1 models at https://huggingface.co/models?filter=mobilenet_v1
+]
+
+
+def _build_tf_to_pytorch_map(model, config, tf_weights=None):
+    """
+    A map of modules from TF to PyTorch.
+    """
+
+    tf_to_pt_map = {}
+
+    if isinstance(model, MobileNetV1ForImageClassification):
+        backbone = model.mobilenet_v1
+    else:
+        backbone = model
+
+    prefix = "MobilenetV1/Conv2d_0/"
+    tf_to_pt_map[prefix + "weights"] = backbone.conv_stem.convolution.weight
+    tf_to_pt_map[prefix + "BatchNorm/beta"] = backbone.conv_stem.normalization.bias
+    tf_to_pt_map[prefix + "BatchNorm/gamma"] = backbone.conv_stem.normalization.weight
+    tf_to_pt_map[prefix + "BatchNorm/moving_mean"] = backbone.conv_stem.normalization.running_mean
+    tf_to_pt_map[prefix + "BatchNorm/moving_variance"] = backbone.conv_stem.normalization.running_var
+
+    for i in range(13):
+        tf_index = i + 1
+        pt_index = i * 2
+
+        pointer = backbone.layer[pt_index]
+        prefix = f"MobilenetV1/Conv2d_{tf_index}_depthwise/"
+        tf_to_pt_map[prefix + "depthwise_weights"] = pointer.convolution.weight
+        tf_to_pt_map[prefix + "BatchNorm/beta"] = pointer.normalization.bias
+        tf_to_pt_map[prefix + "BatchNorm/gamma"] = pointer.normalization.weight
+        tf_to_pt_map[prefix + "BatchNorm/moving_mean"] = pointer.normalization.running_mean
+        tf_to_pt_map[prefix + "BatchNorm/moving_variance"] = pointer.normalization.running_var
+
+        pointer = backbone.layer[pt_index + 1]
+        prefix = f"MobilenetV1/Conv2d_{tf_index}_pointwise/"
+        tf_to_pt_map[prefix + "weights"] = pointer.convolution.weight
+        tf_to_pt_map[prefix + "BatchNorm/beta"] = pointer.normalization.bias
+        tf_to_pt_map[prefix + "BatchNorm/gamma"] = pointer.normalization.weight
+        tf_to_pt_map[prefix + "BatchNorm/moving_mean"] = pointer.normalization.running_mean
+        tf_to_pt_map[prefix + "BatchNorm/moving_variance"] = pointer.normalization.running_var
+
+    if isinstance(model, MobileNetV1ForImageClassification):
+        prefix = "MobilenetV1/Logits/Conv2d_1c_1x1/"
+        tf_to_pt_map[prefix + "weights"] = model.classifier.weight
+        tf_to_pt_map[prefix + "biases"] = model.classifier.bias
+
+    return tf_to_pt_map
+
+
+def load_tf_weights_in_mobilenet_v1(model, config, tf_checkpoint_path):
+    """Load TensorFlow checkpoints in a PyTorch model."""
+    try:
+        import numpy as np
+        import tensorflow as tf
+    except ImportError:
+        logger.error(
+            "Loading a TensorFlow models in PyTorch, requires TensorFlow to be installed. Please see "
+            "https://www.tensorflow.org/install/ for installation instructions."
+        )
+        raise
+
+    # Load weights from TF model
+    init_vars = tf.train.list_variables(tf_checkpoint_path)
+    tf_weights = {}
+    for name, shape in init_vars:
+        logger.info(f"Loading TF weight {name} with shape {shape}")
+        array = tf.train.load_variable(tf_checkpoint_path, name)
+        tf_weights[name] = array
+
+    # Build TF to PyTorch weights loading map
+    tf_to_pt_map = _build_tf_to_pytorch_map(model, config, tf_weights)
+
+    for name, pointer in tf_to_pt_map.items():
+        logger.info(f"Importing {name}")
+        if name not in tf_weights:
+            logger.info(f"{name} not in tf pre-trained weights, skipping")
+            continue
+
+        array = tf_weights[name]
+
+        if "depthwise_weights" in name:
+            logger.info("Transposing depthwise")
+            array = np.transpose(array, (2, 3, 0, 1))
+        elif "weights" in name:
+            logger.info("Transposing")
+            if len(pointer.shape) == 2:  # copying into linear layer
+                array = array.squeeze().transpose()
+            else:
+                array = np.transpose(array, (3, 2, 0, 1))
+
+        if pointer.shape != array.shape:
+            raise ValueError(f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched")
+
+        logger.info(f"Initialize PyTorch weight {name} {array.shape}")
+        pointer.data = torch.from_numpy(array)
+
+        tf_weights.pop(name, None)
+        tf_weights.pop(name + "/RMSProp", None)
+        tf_weights.pop(name + "/RMSProp_1", None)
+        tf_weights.pop(name + "/ExponentialMovingAverage", None)
+
+    logger.info(f"Weights not copied to PyTorch model: {', '.join(tf_weights.keys())}")
+    return model
+
+
+def apply_tf_padding(features: torch.Tensor, conv_layer: nn.Conv2d) -> torch.Tensor:
+    """
+    Apply TensorFlow-style "SAME" padding to a convolution layer. See the notes at:
+    https://www.tensorflow.org/api_docs/python/tf/nn#notes_on_padding_2
+    """
+    in_height, in_width = features.shape[-2:]
+    stride_height, stride_width = conv_layer.stride
+    kernel_height, kernel_width = conv_layer.kernel_size
+
+    if in_height % stride_height == 0:
+        pad_along_height = max(kernel_height - stride_height, 0)
+    else:
+        pad_along_height = max(kernel_height - (in_height % stride_height), 0)
+
+    if in_width % stride_width == 0:
+        pad_along_width = max(kernel_width - stride_width, 0)
+    else:
+        pad_along_width = max(kernel_width - (in_width % stride_width), 0)
+
+    pad_left = pad_along_width // 2
+    pad_right = pad_along_width - pad_left
+    pad_top = pad_along_height // 2
+    pad_bottom = pad_along_height - pad_top
+
+    padding = (pad_left, pad_right, pad_top, pad_bottom)
+    return nn.functional.pad(features, padding, "constant", 0.0)
+
+
+class MobileNetV1ConvLayer(nn.Module):
+    def __init__(
+        self,
+        config: MobileNetV1Config,
+        in_channels: int,
+        out_channels: int,
+        kernel_size: int,
+        stride: Optional[int] = 1,
+        groups: Optional[int] = 1,
+        bias: bool = False,
+        use_normalization: Optional[bool] = True,
+        use_activation: Optional[bool or str] = True,
+    ) -> None:
+        super().__init__()
+        self.config = config
+
+        if in_channels % groups != 0:
+            raise ValueError(f"Input channels ({in_channels}) are not divisible by {groups} groups.")
+        if out_channels % groups != 0:
+            raise ValueError(f"Output channels ({out_channels}) are not divisible by {groups} groups.")
+
+        padding = 0 if config.tf_padding else int((kernel_size - 1) / 2)
+
+        self.convolution = nn.Conv2d(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            bias=bias,
+            padding_mode="zeros",
+        )
+
+        if use_normalization:
+            self.normalization = nn.BatchNorm2d(
+                num_features=out_channels,
+                eps=config.layer_norm_eps,
+                momentum=0.9997,
+                affine=True,
+                track_running_stats=True,
+            )
+        else:
+            self.normalization = None
+
+        if use_activation:
+            if isinstance(use_activation, str):
+                self.activation = ACT2FN[use_activation]
+            elif isinstance(config.hidden_act, str):
+                self.activation = ACT2FN[config.hidden_act]
+            else:
+                self.activation = config.hidden_act
+        else:
+            self.activation = None
+
+    def forward(self, features: torch.Tensor) -> torch.Tensor:
+        if self.config.tf_padding:
+            features = apply_tf_padding(features, self.convolution)
+        features = self.convolution(features)
+        if self.normalization is not None:
+            features = self.normalization(features)
+        if self.activation is not None:
+            features = self.activation(features)
+        return features
+
+
+class MobileNetV1PreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = MobileNetV1Config
+    load_tf_weights = load_tf_weights_in_mobilenet_v1
+    base_model_prefix = "mobilenet_v1"
+    main_input_name = "pixel_values"
+    supports_gradient_checkpointing = False
+
+    def _init_weights(self, module: Union[nn.Linear, nn.Conv2d]) -> None:
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.BatchNorm2d):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+
+MOBILENET_V1_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it
+    as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config ([`MobileNetV1Config`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+MOBILENET_V1_INPUTS_DOCSTRING = r"""
+    Args:
+        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
+            Pixel values. Pixel values can be obtained using [`MobileNetV1FeatureExtractor`]. See
+            [`MobileNetV1FeatureExtractor.__call__`] for details.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare MobileNetV1 model outputting raw hidden-states without any specific head on top.",
+    MOBILENET_V1_START_DOCSTRING,
+)
+class MobileNetV1Model(MobileNetV1PreTrainedModel):
+    def __init__(self, config: MobileNetV1Config, add_pooling_layer: bool = True):
+        super().__init__(config)
+        self.config = config
+
+        depth = 32
+        out_channels = max(int(depth * config.depth_multiplier), config.min_depth)
+
+        self.conv_stem = MobileNetV1ConvLayer(
+            config,
+            in_channels=config.num_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=2,
+        )
+
+        strides = [1, 2, 1, 2, 1, 2, 1, 1, 1, 1, 1, 2, 1]
+
+        self.layer = nn.ModuleList()
+        for i in range(13):
+            in_channels = out_channels
+
+            if strides[i] == 2 or i == 0:
+                depth *= 2
+                out_channels = max(int(depth * config.depth_multiplier), config.min_depth)
+
+            self.layer.append(
+                MobileNetV1ConvLayer(
+                    config,
+                    in_channels=in_channels,
+                    out_channels=in_channels,
+                    kernel_size=3,
+                    stride=strides[i],
+                    groups=in_channels,
+                )
+            )
+
+            self.layer.append(
+                MobileNetV1ConvLayer(
+                    config,
+                    in_channels=in_channels,
+                    out_channels=out_channels,
+                    kernel_size=1,
+                )
+            )
+
+        self.pooler = nn.AdaptiveAvgPool2d((1, 1)) if add_pooling_layer else None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def _prune_heads(self, heads_to_prune):
+        raise NotImplementedError
+
+    @add_start_docstrings_to_model_forward(MOBILENET_V1_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        processor_class=_FEAT_EXTRACTOR_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPoolingAndNoAttention,
+        config_class=_CONFIG_FOR_DOC,
+        modality="vision",
+        expected_output=_EXPECTED_OUTPUT_SHAPE,
+    )
+    def forward(
+        self,
+        pixel_values: Optional[torch.Tensor] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[tuple, BaseModelOutputWithPoolingAndNoAttention]:
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if pixel_values is None:
+            raise ValueError("You have to specify pixel_values")
+
+        hidden_states = self.conv_stem(pixel_values)
+
+        all_hidden_states = () if output_hidden_states else None
+
+        for i, layer_module in enumerate(self.layer):
+            hidden_states = layer_module(hidden_states)
+
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+        last_hidden_state = hidden_states
+
+        if self.pooler is not None:
+            pooled_output = torch.flatten(self.pooler(last_hidden_state), start_dim=1)
+        else:
+            pooled_output = None
+
+        if not return_dict:
+            return tuple(v for v in [last_hidden_state, pooled_output, all_hidden_states] if v is not None)
+
+        return BaseModelOutputWithPoolingAndNoAttention(
+            last_hidden_state=last_hidden_state,
+            pooler_output=pooled_output,
+            hidden_states=all_hidden_states,
+        )
+
+
+@add_start_docstrings(
+    """
+    MobileNetV1 model with an image classification head on top (a linear layer on top of the pooled features), e.g. for
+    ImageNet.
+    """,
+    MOBILENET_V1_START_DOCSTRING,
+)
+class MobileNetV1ForImageClassification(MobileNetV1PreTrainedModel):
+    def __init__(self, config: MobileNetV1Config) -> None:
+        super().__init__(config)
+
+        self.num_labels = config.num_labels
+        self.mobilenet_v1 = MobileNetV1Model(config)
+
+        last_hidden_size = self.mobilenet_v1.layer[-1].convolution.out_channels
+
+        # Classifier head
+        self.dropout = nn.Dropout(config.classifier_dropout_prob, inplace=True)
+        self.classifier = nn.Linear(last_hidden_size, config.num_labels) if config.num_labels > 0 else nn.Identity()
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(MOBILENET_V1_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        processor_class=_FEAT_EXTRACTOR_FOR_DOC,
+        checkpoint=_IMAGE_CLASS_CHECKPOINT,
+        output_type=ImageClassifierOutputWithNoAttention,
+        config_class=_CONFIG_FOR_DOC,
+        expected_output=_IMAGE_CLASS_EXPECTED_OUTPUT,
+    )
+    def forward(
+        self,
+        pixel_values: Optional[torch.Tensor] = None,
+        output_hidden_states: Optional[bool] = None,
+        labels: Optional[torch.Tensor] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[tuple, ImageClassifierOutputWithNoAttention]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the image classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss). If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.mobilenet_v1(pixel_values, output_hidden_states=output_hidden_states, return_dict=return_dict)
+
+        pooled_output = outputs.pooler_output if return_dict else outputs[1]
+
+        logits = self.classifier(self.dropout(pooled_output))
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return ImageClassifierOutputWithNoAttention(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+        )

src/transformers/models/mobilenet_v2/configuration_mobilenet_v2.py

@@ -27,10 +27,10 @@ from ...utils import logging
 logger = logging.get_logger(__name__)
 
 MOBILENET_V2_PRETRAINED_CONFIG_ARCHIVE_MAP = {
-    "google/mobilenet_v2_1.4_224": "https://huggingface.co/Matthijs/mobilenet_v2_1.4_224/resolve/main/config.json",
-    "google/mobilenet_v2_1.0_224": "https://huggingface.co/Matthijs/mobilenet_v2_1.0_224/resolve/main/config.json",
-    "google/mobilenet_v2_0.75_160": "https://huggingface.co/Matthijs/mobilenet_v2_0.75_160/resolve/main/config.json",
-    "google/mobilenet_v2_0.35_96": "https://huggingface.co/Matthijs/mobilenet_v2_0.35_96/resolve/main/config.json",
+    "google/mobilenet_v2_1.4_224": "https://huggingface.co/google/mobilenet_v2_1.4_224/resolve/main/config.json",
+    "google/mobilenet_v2_1.0_224": "https://huggingface.co/google/mobilenet_v2_1.0_224/resolve/main/config.json",
+    "google/mobilenet_v2_0.75_160": "https://huggingface.co/google/mobilenet_v2_0.75_160/resolve/main/config.json",
+    "google/mobilenet_v2_0.35_96": "https://huggingface.co/google/mobilenet_v2_0.35_96/resolve/main/config.json",
     # See all MobileNetV2 models at https://huggingface.co/models?filter=mobilenet_v2
 }
 
@@ -40,7 +40,7 @@ class MobileNetV2Config(PretrainedConfig):
     This is the configuration class to store the configuration of a [`MobileNetV2Model`]. It is used to instantiate a
     MobileNetV2 model according to the specified arguments, defining the model architecture. Instantiating a
     configuration with the defaults will yield a similar configuration to that of the MobileNetV2
-    [Matthijs/mobilenet_v2_1.0_224](https://huggingface.co/Matthijs/mobilenet_v2_1.0_224) architecture.
+    [google/mobilenet_v2_1.0_224](https://huggingface.co/google/mobilenet_v2_1.0_224) architecture.
 
     Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
     documentation from [`PretrainedConfig`] for more information.

src/transformers/onnx/features.py

@@ -408,6 +408,11 @@ class FeaturesManager:
             "question-answering",
             onnx_config_cls="models.mobilebert.MobileBertOnnxConfig",
         ),
+        "mobilenet_v1": supported_features_mapping(
+            "default",
+            "image-classification",
+            onnx_config_cls="models.mobilenet_v1.MobileNetV1OnnxConfig",
+        ),
         "mobilenet_v2": supported_features_mapping(
             "default",
             "image-classification",

src/transformers/utils/dummy_pt_objects.py

@@ -3571,6 +3571,34 @@ def load_tf_weights_in_mobilebert(*args, **kwargs):
     requires_backends(load_tf_weights_in_mobilebert, ["torch"])
 
 
+MOBILENET_V1_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class MobileNetV1ForImageClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileNetV1Model(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileNetV1PreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+def load_tf_weights_in_mobilenet_v1(*args, **kwargs):
+    requires_backends(load_tf_weights_in_mobilenet_v1, ["torch"])
+
+
 MOBILENET_V2_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 MobileNetV1FeatureExtractor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
+class MobileNetV1ImageProcessor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
 class MobileNetV2FeatureExtractor(metaclass=DummyObject):
     _backends = ["vision"]
 

tests/models/mobilenet_v1/test_feature_extraction_mobilenet_v1.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 MobileNetV1FeatureExtractor
+
+
+class MobileNetV1FeatureExtractionTester(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 MobileNetV1FeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCase):
+
+    feature_extraction_class = MobileNetV1FeatureExtractor if is_vision_available() else None
+
+    def setUp(self):
+        self.feature_extract_tester = MobileNetV1FeatureExtractionTester(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_v1/test_modeling_mobilenet_v1.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 MobileNetV1 model. """
+
+
+import inspect
+import unittest
+
+from transformers import MobileNetV1Config
+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 MobileNetV1ForImageClassification, MobileNetV1Model
+    from transformers.models.mobilenet_v1.modeling_mobilenet_v1 import MOBILENET_V1_PRETRAINED_MODEL_ARCHIVE_LIST
+
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import MobileNetV1FeatureExtractor
+
+
+class MobileNetV1ConfigTester(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 MobileNetV1ModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        num_channels=3,
+        image_size=32,
+        depth_multiplier=0.25,
+        min_depth=8,
+        tf_padding=True,
+        last_hidden_size=1024,
+        output_stride=32,
+        hidden_act="relu6",
+        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.min_depth = min_depth
+        self.tf_padding = tf_padding
+        self.last_hidden_size = int(last_hidden_size * depth_multiplier)
+        self.output_stride = output_stride
+        self.hidden_act = hidden_act
+        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 MobileNetV1Config(
+            num_channels=self.num_channels,
+            image_size=self.image_size,
+            depth_multiplier=self.depth_multiplier,
+            min_depth=self.min_depth,
+            tf_padding=self.tf_padding,
+            hidden_act=self.hidden_act,
+            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 = MobileNetV1Model(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,
+            ),
+        )
+
+    def create_and_check_for_image_classification(self, config, pixel_values, labels, pixel_labels):
+        config.num_labels = self.num_labels
+        model = MobileNetV1ForImageClassification(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 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 MobileNetV1ModelTest(ModelTesterMixin, unittest.TestCase):
+    """
+    Here we also overwrite some of the tests of test_modeling_common.py, as MobileNetV1 does not use input_ids, inputs_embeds,
+    attention_mask and seq_length.
+    """
+
+    all_model_classes = (MobileNetV1Model, MobileNetV1ForImageClassification) 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 = MobileNetV1ModelTester(self)
+        self.config_tester = MobileNetV1ConfigTester(self, config_class=MobileNetV1Config, has_text_modality=False)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    @unittest.skip(reason="MobileNetV1 does not use inputs_embeds")
+    def test_inputs_embeds(self):
+        pass
+
+    @unittest.skip(reason="MobileNetV1 does not support input and output embeddings")
+    def test_model_common_attributes(self):
+        pass
+
+    @unittest.skip(reason="MobileNetV1 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 = 26
+            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)
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in MOBILENET_V1_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = MobileNetV1Model.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 MobileNetV1ModelIntegrationTest(unittest.TestCase):
+    @cached_property
+    def default_feature_extractor(self):
+        return (
+            MobileNetV1FeatureExtractor.from_pretrained("google/mobilenet_v1_1.0_224")
+            if is_vision_available()
+            else None
+        )
+
+    @slow
+    def test_inference_image_classification_head(self):
+        model = MobileNetV1ForImageClassification.from_pretrained("google/mobilenet_v1_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([-4.1739, -1.1233, 3.1205]).to(torch_device)
+
+        self.assertTrue(torch.allclose(outputs.logits[0, :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_v1", "google/mobilenet_v1_0.75_192"),
     ("mobilenet_v2", "google/mobilenet_v2_0.35_96"),
     ("mobilevit", "apple/mobilevit-small"),
     ("xlm", "xlm-clm-ende-1024"),

utils/documentation_tests.txt

@@ -113,6 +113,7 @@ 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_v1/modeling_mobilenet_v1.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