pytorch 加载不完全相同结构的模型到当前模型结构

在魔改模型后还想要加载原先能加载的部分训练权重

     # 加载预训练权重
        pretrained_dict = torch.load(pretrained_path)
        if 'state_dict' in pretrained_dict:
            pretrained_dict = pretrained_dict['state_dict']
        
        # 过滤掉不匹配的键
        model_dict = self.resnet.state_dict()
        pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict and v.size() == model_dict[k].size()}
        model_dict.update(pretrained_dict)
        self.resnet.load_state_dict(model_dict)

`