File size: 15,726 Bytes
34a4bcb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) MONAI Consortium
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#     http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from __future__ import annotations

import math
from collections.abc import Sequence

import torch
from torch import nn

from monai.config.type_definitions import PathLike
from monai.utils import optional_import

transformers = optional_import("transformers")
load_tf_weights_in_bert = optional_import("transformers", name="load_tf_weights_in_bert")[0]
cached_file = optional_import("transformers.utils", name="cached_file")[0]
BertEmbeddings = optional_import("transformers.models.bert.modeling_bert", name="BertEmbeddings")[0]
BertLayer = optional_import("transformers.models.bert.modeling_bert", name="BertLayer")[0]

__all__ = ["BertPreTrainedModel", "BertAttention", "BertOutput", "BertMixedLayer", "Pooler", "MultiModal", "Transchex"]


class BertPreTrainedModel(nn.Module):
    """Module to load BERT pre-trained weights.
    Based on:
    LXMERT
    https://github.com/airsplay/lxmert
    BERT (pytorch-transformer)
    https://github.com/huggingface/transformers
    """

    def __init__(self, *inputs, **kwargs) -> None:
        super().__init__()

    def init_bert_weights(self, module):
        if isinstance(module, (nn.Linear, nn.Embedding)):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
        elif isinstance(module, torch.nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)
        if isinstance(module, nn.Linear) and module.bias is not None:
            module.bias.data.zero_()

    @classmethod
    def from_pretrained(
        cls,
        num_language_layers,
        num_vision_layers,
        num_mixed_layers,
        bert_config,
        state_dict=None,
        cache_dir=None,
        from_tf=False,
        path_or_repo_id="bert-base-uncased",
        filename="pytorch_model.bin",
        *inputs,
        **kwargs,
    ):
        weights_path = cached_file(path_or_repo_id, filename, cache_dir=cache_dir)
        model = cls(num_language_layers, num_vision_layers, num_mixed_layers, bert_config, *inputs, **kwargs)
        if state_dict is None and not from_tf:
            state_dict = torch.load(weights_path, map_location="cpu" if not torch.cuda.is_available() else None)
        if from_tf:
            return load_tf_weights_in_bert(model, weights_path)
        old_keys = []
        new_keys = []
        for key in state_dict.keys():
            new_key = None
            if "gamma" in key:
                new_key = key.replace("gamma", "weight")
            if "beta" in key:
                new_key = key.replace("beta", "bias")
            if new_key:
                old_keys.append(key)
                new_keys.append(new_key)
        for old_key, new_key in zip(old_keys, new_keys):
            state_dict[new_key] = state_dict.pop(old_key)
        missing_keys: list = []
        unexpected_keys: list = []
        error_msgs: list = []
        metadata = getattr(state_dict, "_metadata", None)
        state_dict = state_dict.copy()
        if metadata is not None:
            state_dict._metadata = metadata

        def load(module, prefix=""):
            local_metadata = {} if metadata is None else metadata.get(prefix[:-1], {})
            module._load_from_state_dict(
                state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs
            )
            for name, child in module._modules.items():
                if child is not None:
                    load(child, prefix + name + ".")

        start_prefix = ""
        if not hasattr(model, "bert") and any(s.startswith("bert.") for s in state_dict.keys()):
            start_prefix = "bert."
        load(model, prefix=start_prefix)
        return model


class BertAttention(nn.Module):
    """BERT attention layer.
    Based on: BERT (pytorch-transformer)
    https://github.com/huggingface/transformers
    """

    def __init__(self, config) -> None:
        super().__init__()
        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size
        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        self.key = nn.Linear(config.hidden_size, self.all_head_size)
        self.value = nn.Linear(config.hidden_size, self.all_head_size)
        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(self, hidden_states, context):
        mixed_query_layer = self.query(hidden_states)
        mixed_key_layer = self.key(context)
        mixed_value_layer = self.value(context)
        query_layer = self.transpose_for_scores(mixed_query_layer)
        key_layer = self.transpose_for_scores(mixed_key_layer)
        value_layer = self.transpose_for_scores(mixed_value_layer)
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        attention_probs = self.dropout(nn.Softmax(dim=-1)(attention_scores))
        context_layer = torch.matmul(attention_probs, value_layer)
        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)
        return context_layer


class BertOutput(nn.Module):
    """BERT output layer.
    Based on: BERT (pytorch-transformer)
    https://github.com/huggingface/transformers
    """

    def __init__(self, config) -> None:
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = torch.nn.LayerNorm(config.hidden_size, eps=1e-12)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class BertMixedLayer(nn.Module):
    """BERT cross attention layer.
    Based on: BERT (pytorch-transformer)
    https://github.com/huggingface/transformers
    """

    def __init__(self, config) -> None:
        super().__init__()
        self.att_x = BertAttention(config)
        self.output_x = BertOutput(config)
        self.att_y = BertAttention(config)
        self.output_y = BertOutput(config)

    def forward(self, x, y):
        output_x = self.att_x(x, y)
        output_y = self.att_y(y, x)
        return self.output_x(output_x, x), self.output_y(output_y, y)


class Pooler(nn.Module):
    """BERT pooler layer.
    Based on: BERT (pytorch-transformer)
    https://github.com/huggingface/transformers
    """

    def __init__(self, hidden_size) -> None:
        super().__init__()
        self.dense = nn.Linear(hidden_size, hidden_size)
        self.activation = nn.Tanh()

    def forward(self, hidden_states):
        first_token_tensor = hidden_states[:, 0]
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output


class MultiModal(BertPreTrainedModel):
    """
    Multimodal Transformers From Pretrained BERT Weights"
    """

    def __init__(
        self, num_language_layers: int, num_vision_layers: int, num_mixed_layers: int, bert_config: dict
    ) -> None:
        """
        Args:
            num_language_layers: number of language transformer layers.
            num_vision_layers: number of vision transformer layers.
            bert_config: configuration for bert language transformer encoder.

        """
        super().__init__()
        self.config = type("obj", (object,), bert_config)
        self.embeddings = BertEmbeddings(self.config)
        self.language_encoder = nn.ModuleList([BertLayer(self.config) for _ in range(num_language_layers)])
        self.vision_encoder = nn.ModuleList([BertLayer(self.config) for _ in range(num_vision_layers)])
        self.mixed_encoder = nn.ModuleList([BertMixedLayer(self.config) for _ in range(num_mixed_layers)])
        self.apply(self.init_bert_weights)

    def forward(self, input_ids, token_type_ids=None, vision_feats=None, attention_mask=None):
        language_features = self.embeddings(input_ids, token_type_ids)
        for layer in self.vision_encoder:
            vision_feats = layer(vision_feats, None)[0]
        for layer in self.language_encoder:
            language_features = layer(language_features, attention_mask)[0]
        for layer in self.mixed_encoder:
            language_features, vision_feats = layer(language_features, vision_feats)
        return language_features, vision_feats


class Transchex(torch.nn.Module):
    """
    TransChex based on: "Hatamizadeh et al.,TransCheX: Self-Supervised Pretraining of Vision-Language
    Transformers for Chest X-ray Analysis"
    """

    def __init__(
        self,
        in_channels: int,
        img_size: Sequence[int] | int,
        patch_size: int | tuple[int, int],
        num_classes: int,
        num_language_layers: int,
        num_vision_layers: int,
        num_mixed_layers: int,
        hidden_size: int = 768,
        drop_out: float = 0.0,
        attention_probs_dropout_prob: float = 0.1,
        gradient_checkpointing: bool = False,
        hidden_act: str = "gelu",
        hidden_dropout_prob: float = 0.1,
        initializer_range: float = 0.02,
        intermediate_size: int = 3072,
        layer_norm_eps: float = 1e-12,
        max_position_embeddings: int = 512,
        model_type: str = "bert",
        num_attention_heads: int = 12,
        num_hidden_layers: int = 12,
        pad_token_id: int = 0,
        position_embedding_type: str = "absolute",
        transformers_version: str = "4.10.2",
        type_vocab_size: int = 2,
        use_cache: bool = True,
        vocab_size: int = 30522,
        chunk_size_feed_forward: int = 0,
        is_decoder: bool = False,
        add_cross_attention: bool = False,
        path_or_repo_id: str | PathLike = "bert-base-uncased",
        filename: str = "pytorch_model.bin",
    ) -> None:
        """
        Args:
            in_channels: dimension of input channels.
            img_size: dimension of input image.
            patch_size: dimension of patch size.
            num_classes: number of classes if classification is used.
            num_language_layers: number of language transformer layers.
            num_vision_layers: number of vision transformer layers.
            num_mixed_layers: number of mixed transformer layers.
            drop_out: fraction of the input units to drop.
            path_or_repo_id: This can be either:
                - a string, the *model id* of a model repo on huggingface.co.
                - a path to a *directory* potentially containing the file.
            filename: The name of the file to locate in `path_or_repo`.

        The other parameters are part of the `bert_config` to `MultiModal.from_pretrained`.

        Examples:

        .. code-block:: python

            # for 3-channel with image size of (224,224), patch size of (32,32), 3 classes, 2 language layers,
            # 2 vision layers, 2 mixed modality layers and dropout of 0.2 in the classification head
            net = Transchex(in_channels=3,
                                 img_size=(224, 224),
                                 num_classes=3,
                                 num_language_layers=2,
                                 num_vision_layers=2,
                                 num_mixed_layers=2,
                                 drop_out=0.2)

        """
        super().__init__()
        bert_config = {
            "attention_probs_dropout_prob": attention_probs_dropout_prob,
            "classifier_dropout": None,
            "gradient_checkpointing": gradient_checkpointing,
            "hidden_act": hidden_act,
            "hidden_dropout_prob": hidden_dropout_prob,
            "hidden_size": hidden_size,
            "initializer_range": initializer_range,
            "intermediate_size": intermediate_size,
            "layer_norm_eps": layer_norm_eps,
            "max_position_embeddings": max_position_embeddings,
            "model_type": model_type,
            "num_attention_heads": num_attention_heads,
            "num_hidden_layers": num_hidden_layers,
            "pad_token_id": pad_token_id,
            "position_embedding_type": position_embedding_type,
            "transformers_version": transformers_version,
            "type_vocab_size": type_vocab_size,
            "use_cache": use_cache,
            "vocab_size": vocab_size,
            "chunk_size_feed_forward": chunk_size_feed_forward,
            "is_decoder": is_decoder,
            "add_cross_attention": add_cross_attention,
        }
        if not (0 <= drop_out <= 1):
            raise ValueError("dropout_rate should be between 0 and 1.")

        if (img_size[0] % patch_size[0] != 0) or (img_size[1] % patch_size[1] != 0):  # type: ignore
            raise ValueError("img_size should be divisible by patch_size.")

        self.multimodal = MultiModal.from_pretrained(
            num_language_layers=num_language_layers,
            num_vision_layers=num_vision_layers,
            num_mixed_layers=num_mixed_layers,
            bert_config=bert_config,
            path_or_repo_id=path_or_repo_id,
            filename=filename,
        )

        self.patch_size = patch_size
        self.num_patches = (img_size[0] // self.patch_size[0]) * (img_size[1] // self.patch_size[1])  # type: ignore
        self.vision_proj = nn.Conv2d(
            in_channels=in_channels, out_channels=hidden_size, kernel_size=self.patch_size, stride=self.patch_size
        )
        self.norm_vision_pos = nn.LayerNorm(hidden_size)
        self.pos_embed_vis = nn.Parameter(torch.zeros(1, self.num_patches, hidden_size))
        self.pooler = Pooler(hidden_size=hidden_size)
        self.drop = torch.nn.Dropout(drop_out)
        self.cls_head = torch.nn.Linear(hidden_size, num_classes)

    def forward(self, input_ids, token_type_ids=None, vision_feats=None):
        attention_mask = torch.ones_like(input_ids).unsqueeze(1).unsqueeze(2)
        attention_mask = attention_mask.to(dtype=next(self.parameters()).dtype)
        attention_mask = (1.0 - attention_mask) * -10000.0
        vision_feats = self.vision_proj(vision_feats).flatten(2).transpose(1, 2)
        vision_feats = self.norm_vision_pos(vision_feats)
        vision_feats = vision_feats + self.pos_embed_vis
        hidden_state_lang, hidden_state_vis = self.multimodal(
            input_ids=input_ids, token_type_ids=token_type_ids, vision_feats=vision_feats, attention_mask=attention_mask
        )
        pooled_features = self.pooler(hidden_state_lang)
        logits = self.cls_head(self.drop(pooled_features))
        return logits