在进行模型测试时报错:测试
Missing key(s) in state_dict: xxxxxxxxxx.net
Unexpected key(s) in state_dict:xxxxxxxxxxcode
报错缘由:blog
在模型训练时有加上:【能够加速训练速度】get
model = nn.DataParallel(model)博客
#cudnn.benchmark = Trueio
可是在模型测试推断时,在模型参数被加载到模型前没有加这句话,故报出上面的错误。class
解决:model
在模型参数被加载到模型前加下面的语句:map
model = nn.DataParallel(model)
#cudnn.benchmark = True
同时在加载模型的时候,添加一个False
weight = os.path.join(weight_path) chkpt = torch.load(weight, map_location=self.__device) self.__model.load_state_dict(chkpt, False)
参考博客:
https://blog.csdn.net/u013925378/article/details/104749000/
https://blog.csdn.net/sinat_34054843/article/details/88046041