ERROR: 加载预训练模型参数时出现size mismatch错误

 原来的resnet为：

def resnet152(pretrained=False, **kwargs):
    """Constructs a ResNet-152 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
    if pretrained:
        model.load_state_dict(model_zoo.load_url(model_urls['resnet152']))
    return model

 修改为：

def resnet152(pretrained=False, **kwargs):
    """Constructs a ResNet-152 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
    #if pretrained:
     #   model.load_state_dict(model_zoo.load_url(model_urls['resnet152']))
    return model

再通过以下程序加载预训练模型：

model = resnet152(pretrained=True, **kwargs)
model_dict = model.state_dict()
state_dict = torch.load('resnet152-b121ed2d.pth')
new_state_dict = {k: v for k, v in state_dict.items() if k in model_dict}
model_dict.update(new_state_dict)
model.load_state_dict(model_dict)

或者直接尝试如下方法：

model.load_state_dict(torch.load("model.th"),strict=False)

将strict设置为False