pytorch 预训练model加载问题”Unexpected key(s) in state_dict:“

预训练模型使用

    model = ResNet50()
    model = nn.DataParallel(model).cuda()

加载预训练模型是使用
 

model = ResNet50().cuda()

checkpoint = torch.load(args.net_cache)

model.load_state_dict(checkpoint['state_dict'])//

报错：RuntimeError: Error(s) in loading state_dict for Resnet50   Unexpected key(s) in state_dict:““

说明加载模型和预训练模型环境不一致，本来修改如下(后经权重输出对比发现这种方式加载的model与原model权重不一致)：

model = ResNet50().cuda()

checkpoint = torch.load(args.net_cache)

model.load_state_dict(checkpoint['state_dict'],False)//

model.load_state_dict(state_dict, strict=True)

解决1：由于用DataParallel训练的模型数据并行方式的，key中会包含”module“关键字,加载时直接用:

model = ResNet50().cuda()
model = nn.DataParallel(model)

checkpoint = t