File size: 22,179 Bytes
1ce5e18
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
# coding=utf-8
# Copyright 2022 Meta Platforms, 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 ConvNext model."""


from typing import Optional, Tuple, Union

import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss

from ...activations import ACT2FN
from ...modeling_outputs import (
    BackboneOutput,
    BaseModelOutputWithNoAttention,
    BaseModelOutputWithPoolingAndNoAttention,
    ImageClassifierOutputWithNoAttention,
)
from ...modeling_utils import PreTrainedModel
from ...utils import (
    add_code_sample_docstrings,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    logging,
    replace_return_docstrings,
)
from ...utils.backbone_utils import BackboneMixin
from .configuration_convnext import ConvNextConfig


logger = logging.get_logger(__name__)

# General docstring
_CONFIG_FOR_DOC = "ConvNextConfig"

# Base docstring
_CHECKPOINT_FOR_DOC = "facebook/convnext-tiny-224"
_EXPECTED_OUTPUT_SHAPE = [1, 768, 7, 7]

# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "facebook/convnext-tiny-224"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"

CONVNEXT_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "facebook/convnext-tiny-224",
    # See all ConvNext models at https://huggingface.co/models?filter=convnext
]


# Copied from transformers.models.beit.modeling_beit.drop_path
def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
    """
    Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

    Comment by Ross Wightman: This is the same as the DropConnect impl I created for EfficientNet, etc networks,
    however, the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for changing the
    layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use 'survival rate' as the
    argument.
    """
    if drop_prob == 0.0 or not training:
        return input
    keep_prob = 1 - drop_prob
    shape = (input.shape[0],) + (1,) * (input.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = keep_prob + torch.rand(shape, dtype=input.dtype, device=input.device)
    random_tensor.floor_()  # binarize
    output = input.div(keep_prob) * random_tensor
    return output


# Copied from transformers.models.beit.modeling_beit.BeitDropPath with Beit->ConvNext
class ConvNextDropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks)."""

    def __init__(self, drop_prob: Optional[float] = None) -> None:
        super().__init__()
        self.drop_prob = drop_prob

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        return drop_path(hidden_states, self.drop_prob, self.training)

    def extra_repr(self) -> str:
        return "p={}".format(self.drop_prob)


class ConvNextLayerNorm(nn.Module):
    r"""LayerNorm that supports two data formats: channels_last (default) or channels_first.
    The ordering of the dimensions in the inputs. channels_last corresponds to inputs with shape (batch_size, height,
    width, channels) while channels_first corresponds to inputs with shape (batch_size, channels, height, width).
    """

    def __init__(self, normalized_shape, eps=1e-6, data_format="channels_last"):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(normalized_shape))
        self.bias = nn.Parameter(torch.zeros(normalized_shape))
        self.eps = eps
        self.data_format = data_format
        if self.data_format not in ["channels_last", "channels_first"]:
            raise NotImplementedError(f"Unsupported data format: {self.data_format}")
        self.normalized_shape = (normalized_shape,)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        if self.data_format == "channels_last":
            x = torch.nn.functional.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
        elif self.data_format == "channels_first":
            input_dtype = x.dtype
            x = x.float()
            u = x.mean(1, keepdim=True)
            s = (x - u).pow(2).mean(1, keepdim=True)
            x = (x - u) / torch.sqrt(s + self.eps)
            x = x.to(dtype=input_dtype)
            x = self.weight[:, None, None] * x + self.bias[:, None, None]
        return x


class ConvNextEmbeddings(nn.Module):
    """This class is comparable to (and inspired by) the SwinEmbeddings class
    found in src/transformers/models/swin/modeling_swin.py.
    """

    def __init__(self, config):
        super().__init__()
        self.patch_embeddings = nn.Conv2d(
            config.num_channels, config.hidden_sizes[0], kernel_size=config.patch_size, stride=config.patch_size
        )
        self.layernorm = ConvNextLayerNorm(config.hidden_sizes[0], eps=1e-6, data_format="channels_first")
        self.num_channels = config.num_channels

    def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
        num_channels = pixel_values.shape[1]
        if num_channels != self.num_channels:
            raise ValueError(
                "Make sure that the channel dimension of the pixel values match with the one set in the configuration."
            )
        embeddings = self.patch_embeddings(pixel_values)
        embeddings = self.layernorm(embeddings)
        return embeddings


class ConvNextLayer(nn.Module):
    """This corresponds to the `Block` class in the original implementation.

    There are two equivalent implementations: [DwConv, LayerNorm (channels_first), Conv, GELU,1x1 Conv]; all in (N, C,
    H, W) (2) [DwConv, Permute to (N, H, W, C), LayerNorm (channels_last), Linear, GELU, Linear]; Permute back

    The authors used (2) as they find it slightly faster in PyTorch.

    Args:
        config ([`ConvNextConfig`]): Model configuration class.
        dim (`int`): Number of input channels.
        drop_path (`float`): Stochastic depth rate. Default: 0.0.
    """

    def __init__(self, config, dim, drop_path=0):
        super().__init__()
        self.dwconv = nn.Conv2d(dim, dim, kernel_size=7, padding=3, groups=dim)  # depthwise conv
        self.layernorm = ConvNextLayerNorm(dim, eps=1e-6)
        self.pwconv1 = nn.Linear(dim, 4 * dim)  # pointwise/1x1 convs, implemented with linear layers
        self.act = ACT2FN[config.hidden_act]
        self.pwconv2 = nn.Linear(4 * dim, dim)
        self.layer_scale_parameter = (
            nn.Parameter(config.layer_scale_init_value * torch.ones((dim)), requires_grad=True)
            if config.layer_scale_init_value > 0
            else None
        )
        self.drop_path = ConvNextDropPath(drop_path) if drop_path > 0.0 else nn.Identity()

    def forward(self, hidden_states: torch.FloatTensor) -> torch.Tensor:
        input = hidden_states
        x = self.dwconv(hidden_states)
        x = x.permute(0, 2, 3, 1)  # (N, C, H, W) -> (N, H, W, C)
        x = self.layernorm(x)
        x = self.pwconv1(x)
        x = self.act(x)
        x = self.pwconv2(x)
        if self.layer_scale_parameter is not None:
            x = self.layer_scale_parameter * x
        x = x.permute(0, 3, 1, 2)  # (N, H, W, C) -> (N, C, H, W)

        x = input + self.drop_path(x)
        return x


class ConvNextStage(nn.Module):
    """ConvNeXT stage, consisting of an optional downsampling layer + multiple residual blocks.

    Args:
        config ([`ConvNextConfig`]): Model configuration class.
        in_channels (`int`): Number of input channels.
        out_channels (`int`): Number of output channels.
        depth (`int`): Number of residual blocks.
        drop_path_rates(`List[float]`): Stochastic depth rates for each layer.
    """

    def __init__(self, config, in_channels, out_channels, kernel_size=2, stride=2, depth=2, drop_path_rates=None):
        super().__init__()

        if in_channels != out_channels or stride > 1:
            self.downsampling_layer = nn.Sequential(
                ConvNextLayerNorm(in_channels, eps=1e-6, data_format="channels_first"),
                nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride),
            )
        else:
            self.downsampling_layer = nn.Identity()
        drop_path_rates = drop_path_rates or [0.0] * depth
        self.layers = nn.Sequential(
            *[ConvNextLayer(config, dim=out_channels, drop_path=drop_path_rates[j]) for j in range(depth)]
        )

    def forward(self, hidden_states: torch.FloatTensor) -> torch.Tensor:
        hidden_states = self.downsampling_layer(hidden_states)
        hidden_states = self.layers(hidden_states)
        return hidden_states


class ConvNextEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.stages = nn.ModuleList()
        drop_path_rates = [
            x.tolist() for x in torch.linspace(0, config.drop_path_rate, sum(config.depths)).split(config.depths)
        ]
        prev_chs = config.hidden_sizes[0]
        for i in range(config.num_stages):
            out_chs = config.hidden_sizes[i]
            stage = ConvNextStage(
                config,
                in_channels=prev_chs,
                out_channels=out_chs,
                stride=2 if i > 0 else 1,
                depth=config.depths[i],
                drop_path_rates=drop_path_rates[i],
            )
            self.stages.append(stage)
            prev_chs = out_chs

    def forward(
        self,
        hidden_states: torch.FloatTensor,
        output_hidden_states: Optional[bool] = False,
        return_dict: Optional[bool] = True,
    ) -> Union[Tuple, BaseModelOutputWithNoAttention]:
        all_hidden_states = () if output_hidden_states else None

        for i, layer_module in enumerate(self.stages):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            hidden_states = layer_module(hidden_states)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, all_hidden_states] if v is not None)

        return BaseModelOutputWithNoAttention(
            last_hidden_state=hidden_states,
            hidden_states=all_hidden_states,
        )


class ConvNextPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = ConvNextConfig
    base_model_prefix = "convnext"
    main_input_name = "pixel_values"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, (nn.Linear, nn.Conv2d)):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            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.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, ConvNextEncoder):
            module.gradient_checkpointing = value


CONVNEXT_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 ([`ConvNextConfig`]): 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.
"""

CONVNEXT_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
            Pixel values. Pixel values can be obtained using [`AutoImageProcessor`]. See
            [`ConvNextImageProcessor.__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 ConvNext model outputting raw features without any specific head on top.",
    CONVNEXT_START_DOCSTRING,
)
class ConvNextModel(ConvNextPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.config = config

        self.embeddings = ConvNextEmbeddings(config)
        self.encoder = ConvNextEncoder(config)

        # final layernorm layer
        self.layernorm = nn.LayerNorm(config.hidden_sizes[-1], eps=config.layer_norm_eps)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(CONVNEXT_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=BaseModelOutputWithPoolingAndNoAttention,
        config_class=_CONFIG_FOR_DOC,
        modality="vision",
        expected_output=_EXPECTED_OUTPUT_SHAPE,
    )
    def forward(
        self,
        pixel_values: torch.FloatTensor = 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")

        embedding_output = self.embeddings(pixel_values)

        encoder_outputs = self.encoder(
            embedding_output,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        last_hidden_state = encoder_outputs[0]

        # global average pooling, (N, C, H, W) -> (N, C)
        pooled_output = self.layernorm(last_hidden_state.mean([-2, -1]))

        if not return_dict:
            return (last_hidden_state, pooled_output) + encoder_outputs[1:]

        return BaseModelOutputWithPoolingAndNoAttention(
            last_hidden_state=last_hidden_state,
            pooler_output=pooled_output,
            hidden_states=encoder_outputs.hidden_states,
        )


@add_start_docstrings(
    """
    ConvNext Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for
    ImageNet.
    """,
    CONVNEXT_START_DOCSTRING,
)
class ConvNextForImageClassification(ConvNextPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.num_labels = config.num_labels
        self.convnext = ConvNextModel(config)

        # Classifier head
        self.classifier = (
            nn.Linear(config.hidden_sizes[-1], 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(CONVNEXT_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        checkpoint=_IMAGE_CLASS_CHECKPOINT,
        output_type=ImageClassifierOutputWithNoAttention,
        config_class=_CONFIG_FOR_DOC,
        expected_output=_IMAGE_CLASS_EXPECTED_OUTPUT,
    )
    def forward(
        self,
        pixel_values: torch.FloatTensor = None,
        labels: Optional[torch.LongTensor] = None,
        output_hidden_states: Optional[bool] = 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.convnext(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(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,
        )


@add_start_docstrings(
    """
    ConvNeXt backbone, to be used with frameworks like DETR and MaskFormer.
    """,
    CONVNEXT_START_DOCSTRING,
)
class ConvNextBackbone(ConvNextPreTrainedModel, BackboneMixin):
    def __init__(self, config):
        super().__init__(config)
        super()._init_backbone(config)

        self.embeddings = ConvNextEmbeddings(config)
        self.encoder = ConvNextEncoder(config)
        self.num_features = [config.hidden_sizes[0]] + config.hidden_sizes

        # Add layer norms to hidden states of out_features
        hidden_states_norms = {}
        for stage, num_channels in zip(self._out_features, self.channels):
            hidden_states_norms[stage] = ConvNextLayerNorm(num_channels, data_format="channels_first")
        self.hidden_states_norms = nn.ModuleDict(hidden_states_norms)

        # initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(CONVNEXT_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=BackboneOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        pixel_values: torch.Tensor,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> BackboneOutput:
        """
        Returns:

        Examples:

        ```python
        >>> from transformers import AutoImageProcessor, AutoBackbone
        >>> import torch
        >>> from PIL import Image
        >>> import requests

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> processor = AutoImageProcessor.from_pretrained("facebook/convnext-tiny-224")
        >>> model = AutoBackbone.from_pretrained("facebook/convnext-tiny-224")

        >>> inputs = processor(image, return_tensors="pt")
        >>> outputs = model(**inputs)
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )

        embedding_output = self.embeddings(pixel_values)

        outputs = self.encoder(
            embedding_output,
            output_hidden_states=True,
            return_dict=True,
        )

        hidden_states = outputs.hidden_states

        feature_maps = ()
        # we skip the stem
        for idx, (stage, hidden_state) in enumerate(zip(self.stage_names[1:], hidden_states[1:])):
            if stage in self.out_features:
                hidden_state = self.hidden_states_norms[stage](hidden_state)
                feature_maps += (hidden_state,)

        if not return_dict:
            output = (feature_maps,)
            if output_hidden_states:
                output += (outputs.hidden_states,)
            return output

        return BackboneOutput(
            feature_maps=feature_maps,
            hidden_states=outputs.hidden_states if output_hidden_states else None,
            attentions=None,
        )