Bert+Resnet联合学习，pytorch模型实例化后为空

Question

我正在写一个简单的联合模型，它有两个分支，一个分支是resnet50，另一个是bert。我连接两个输出并将其传递给具有 2 个输出神经元的简单线性层。

我实现了以下模型：

import torch
from torch import nn
import torchvision.models as models
import torch.nn as nn
from collections import OrderedDict
from transformers import BertModel

class BertResNet(nn.Module):
    def __init__(self):
        super(BertResNet, self).__init__()
        # resnet
        resnet50 = models.resnet50(pretrained=True)
        n_inputs = resnet50.fc.in_features
        # compressed embedding space
        classifier = nn.Sequential(OrderedDict([
            ('fc1', nn.Linear(n_inputs, 512))
        ]))

        resnet50.fc = classifier # 512 out resnet 


        bert = BertModel.from_pretrained('bert-base-uncased')

        # final classification layer

        classification = nn.Linear(512 + 768, 2)
        #print(resnet50)
        #print(bert)

    def forward(self, img, text):
        res_emb = self.resnet50(img)
        bert_emb = self.bert(text)

        combined = torch.cat(res_emb,
                              bet_emb, dim=1)
        out = self.classification(combined)
        return out

但是当我实例化时，我得到一个空模型：

bert_resnet = BertResNet()

print(bert_resnet)

输出： BertResNet()

list(bert_resnet.parameters()) 也 returns []

Answer 1

您从未将模型分配给 BertResNet class 对象的任何属性。 __init__ 方法中有临时变量，但一旦完成，这些变量将被丢弃。他们应该被分配给 self:

def __init__(self):
    super(BertResNet, self).__init__()
    # resnet
    self.resnet50 = models.resnet50(pretrained=True)
    n_inputs = self.resnet50.fc.in_features
    # compressed embedding space
    self.classifier = nn.Sequential(OrderedDict([
        ('fc1', nn.Linear(n_inputs, 512))
    ]))

    self.resnet50.fc = classifier # 512 out resnet 


    self.bert = BertModel.from_pretrained('bert-base-uncased')

    # final classification layer

    self.classification = nn.Linear(512 + 768, 2)

Bert+Resnet联合学习，pytorch模型实例化后为空

Bert + Resnet joint learning, pytorch model is empty after instantiation

pytorch

resnet

bert-language-model

torchvision