File size: 1,628 Bytes
efb665a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn
from transformers import BertModel

class MultimodalClassifier(nn.Module):
    def __init__(self, text_hidden_size=768, image_feat_size=2048, num_classes=5):
        super(MultimodalClassifier, self).__init__()
        self.bert = BertModel.from_pretrained("bert-base-uncased")

        self.text_fc = nn.Sequential(
            nn.Linear(text_hidden_size, 256),
            nn.BatchNorm1d(256),
            nn.ReLU(),
            nn.Dropout(0.2)
        )

        self.image_fc = nn.Sequential(
            nn.Linear(image_feat_size, 256),
            nn.BatchNorm1d(256),
            nn.ReLU(),
            nn.Dropout(0.2)
        )

        self.fusion_fc = nn.Sequential(
            nn.Linear(512, 256),
            nn.ReLU(),
            nn.Dropout(0.3),
            nn.Linear(256, 64),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(64, num_classes)
        )

    def forward(self, input_ids, attention_mask, image_vector):
        text_output = self.bert(input_ids=input_ids, attention_mask=attention_mask)
        text_feat = self.text_fc[0](text_output.pooler_output)
        if text_feat.size(0) > 1:
            text_feat = self.text_fc[1:](text_feat)
        else:
            text_feat = self.text_fc[2:](text_feat)

        image_feat = self.image_fc[0](image_vector)
        if image_feat.size(0) > 1:
            image_feat = self.image_fc[1:](image_feat)
        else:
            image_feat = self.image_fc[2:](image_feat)

        fused = torch.cat((text_feat, image_feat), dim=1)
        logits = self.fusion_fc(fused)
        return logits