问题记录——RuntimeError: Error(s) in loading state_dict for ResNet: Missing key(s) in state_dict：...

项目场景：

在单GPU上调用一个基于多GPU预训练和微调训练的.pth.tar模型文件时，直接加载模型的strict参数:

model.load_state_dict(checkpoint['state_dict'])

这样有可能会出现一个类似的BUG或错误，问题如下

---

问题描述

RuntimeError: Error(s) in loading state_dict for ResNet:
	Missing key(s) in state_dict: "conv1.weight", "bn1.weight", "bn1.bias", "bn1.running_mean", "bn1.running_var", "layer1.0.conv1.weight", "layer1.0.bn1.weight", "layer1.0.bn1.bias", "layer1.0.bn1.running_mean", "layer1.0.bn1.running_var", "layer1.0.conv2.weight", "layer1.0.bn2.weight", "layer1.0.bn2.bias", "layer1.0.bn2.running_mean", "layer1.0.bn2.running_var", "layer1.1.conv1.weight", "layer1.1.bn1.weight", "layer1.1.bn1.bias", "layer1.1.bn1.running_mean", "layer1.1.bn1.running_var", "layer1.1.conv2.weight", "layer1.1.bn2.weight", "layer1.1.bn2.bias", "layer1.1.bn2.running_mean", "layer1.1.bn2.running_var", "layer2.0.conv1.weight", "layer2.0.bn1.weight", "layer2.0.bn1.bias", "layer2.0.bn1.running_mean", "layer2.0.bn1.running_var", "layer2.0.conv2.weight", "layer2.0.bn2.weight", "layer2.0.bn2.bias", "layer2.0.bn2.running_mean", "layer2.0.bn2.running_var", "layer2.0.downsample.0.weight", "layer2.0.downsample.1.weight", "layer2.0.downsample.1.bias", "layer2.0.downsample.1.running_mean", "layer2.0.downsample.1.running_var", "layer2.1.conv1.weight", "layer2.1.bn1.weight", "layer2.1.bn1.bias", "layer2.1.bn1.running_mean", "layer2.1.bn1.running_var", "layer2.1.conv2.weight", "layer2.1.bn2.weight", "layer2.1.bn2.bias", "layer2.1.bn2.running_mean", "layer2.1.bn2.running_var", "layer3.0.conv1.weight", "layer3.0.bn1.weight", "layer3.0.bn1.bias", "layer3.0.bn1.running_mean", "layer3.0.bn1.running_var", "layer3.0.conv2.weight", "layer3.0.bn2.weight", "layer3.0.bn2.bias", "layer3.0.bn2.running_mean", "layer3.0.bn2.running_var", "layer3.0.downsample.0.weight", "layer3.0.downsample.1.weight", "layer3.0.downsample.1.bias", "layer3.0.downsample.1.running_mean", "layer3.0.downsample.1.running_var", "layer3.1.conv1.weight", "layer3.1.bn1.weight", "layer3.1.bn1.bias", "layer3.1.bn1.running_mean", "layer3.1.bn1.running_var", "layer3.1.conv2.weight", "layer3.1.bn2.weight", "layer3.1.bn2.bias", "layer3.1.bn2.running_mean", "layer3.1.bn2.running_var", "layer4.0.conv1.weight", "layer4.0.bn1.weight", "layer4.0.bn1.bias", "layer4.0.bn1.running_mean", "layer4.0.bn1.running_var", "layer4.0.conv2.weight", "layer4.0.bn2.weight", "layer4.0.bn2.bias", "layer4.0.bn2.running_mean", "layer4.0.bn2.running_var", "layer4.0.downsample.0.weight", "layer4.0.downsample.1.weight", "layer4.0.downsample.1.bias", "layer4.0.downsample.1.running_mean", "layer4.0.downsample.1.running_var", "layer4.1.conv1.weight", "layer4.1.bn1.weight", "layer4.1.bn1.bias", "layer4.1.bn1.running_mean", "layer4.1.bn1.running_var", "layer4.1.conv2.weight", "layer4.1.bn2.weight", "layer4.1.bn2.bias", "layer4.1.bn2.running_mean", "layer4.1.bn2.running_var", "fc.weight", "fc.bias". 
	Unexpected key(s) in state_dict: "module.conv1.weight", "module.bn1.weight", "module.bn1.bias", "module.bn1.running_mean", "module.bn1.running_var", "module.bn1.num_batches_tracked", "module.layer1.0.conv1.weight", "module.layer1.0.bn1.weight", "module.layer1.0.bn1.bias", "module.layer1.0.bn1.running_mean", "module.layer1.0.bn1.running_var", "module.layer1.0.bn1.num_batches_tracked", "module.layer1.0.conv2.weight", "module.layer1.0.bn2.weight", "module.layer1.0.bn2.bias", "module.layer1.0.bn2.running_mean", "module.layer1.0.bn2.running_var", "module.layer1.0.bn2.num_batches_tracked", "module.layer1.1.conv1.weight", "module.layer1.1.bn1.weight", "module.layer1.1.bn1.bias", "module.layer1.1.bn1.running_mean", "module.layer1.1.bn1.running_var", "module.layer1.1.bn1.num_batches_tracked", "module.layer1.1.conv2.weight", "module.layer1.1.bn2.weight", "module.layer1.1.bn2.bias", "module.layer1.1.bn2.running_mean", "module.layer1.1.bn2.running_var", "module.layer1.1.bn2.num_batches_tracked", "module.layer2.0.conv1.weight", "module.layer2.0.bn1.weight", "module.layer2.0.bn1.bias", "module.layer2.0.bn1.running_mean", "module.layer2.0.bn1.running_var", "module.layer2.0.bn1.num_batches_tracked", "module.layer2.0.conv2.weight", "module.layer2.0.bn2.weight", "module.layer2.0.bn2.bias", "module.layer2.0.bn2.running_mean", "module.layer2.0.bn2.running_var", "module.layer2.0.bn2.num_batches_tracked", "module.layer2.0.downsample.0.weight", "module.layer2.0.downsample.1.weight", "module.layer2.0.downsample.1.bias", "module.layer2.0.downsample.1.running_mean", "module.layer2.0.downsample.1.running_var", "module.layer2.0.downsample.1.num_batches_tracked", "module.layer2.1.conv1.weight", "module.layer2.1.bn1.weight", "module.layer2.1.bn1.bias", "module.layer2.1.bn1.running_mean", "module.layer2.1.bn1.running_var", "module.layer2.1.bn1.num_batches_tracked", "module.layer2.1.conv2.weight", "module.layer2.1.bn2.weight", "module.layer2.1.bn2.bias", "module.layer2.1.bn2.running_mean", "module.layer2.1.bn2.running_var", "module.layer2.1.bn2.num_batches_tracked", "module.layer3.0.conv1.weight", "module.layer3.0.bn1.weight", "module.layer3.0.bn1.bias", "module.layer3.0.bn1.running_mean", "module.layer3.0.bn1.running_var", "module.layer3.0.bn1.num_batches_tracked", "module.layer3.0.conv2.weight", "module.layer3.0.bn2.weight", "module.layer3.0.bn2.bias", "module.layer3.0.bn2.running_mean", "module.layer3.0.bn2.running_var", "module.layer3.0.bn2.num_batches_tracked", "module.layer3.0.downsample.0.weight", "module.layer3.0.downsample.1.weight", "module.layer3.0.downsample.1.bias", "module.layer3.0.downsample.1.running_mean", "module.layer3.0.downsample.1.running_var", "module.layer3.0.downsample.1.num_batches_tracked", "module.layer3.1.conv1.weight", "module.layer3.1.bn1.weight", "module.layer3.1.bn1.bias", "module.layer3.1.bn1.running_mean", "module.layer3.1.bn1.running_var", "module.layer3.1.bn1.num_batches_tracked", "module.layer3.1.conv2.weight", "module.layer3.1.bn2.weight", "module.layer3.1.bn2.bias", "module.layer3.1.bn2.running_mean", "module.layer3.1.bn2.running_var", "module.layer3.1.bn2.num_batches_tracked", "module.layer4.0.conv1.weight", "module.layer4.0.bn1.weight", "module.layer4.0.bn1.bias", "module.layer4.0.bn1.running_mean", "module.layer4.0.bn1.running_var", "module.layer4.0.bn1.num_batches_tracked", "module.layer4.0.conv2.weight", "module.layer4.0.bn2.weight", "module.layer4.0.bn2.bias", "module.layer4.0.bn2.running_mean", "module.layer4.0.bn2.running_var", "module.layer4.0.bn2.num_batches_tracked", "module.layer4.0.downsample.0.weight", "module.layer4.0.downsample.1.weight", "module.layer4.0.downsample.1.bias", "module.layer4.0.downsample.1.running_mean", "module.layer4.0.downsample.1.running_var", "module.layer4.0.downsample.1.num_batches_tracked", "module.layer4.1.conv1.weight", "module.layer4.1.bn1.weight", "module.layer4.1.bn1.bias", "module.layer4.1.bn1.running_mean", "module.layer4.1.bn1.running_var", "module.layer4.1.bn1.num_batches_tracked", "module.layer4.1.conv2.weight", "module.layer4.1.bn2.weight", "module.layer4.1.bn2.bias", "module.layer4.1.bn2.running_mean", "module.layer4.1.bn2.running_var", "module.layer4.1.bn2.num_batches_tracked", "module.fc.weight", "module.fc.bias". 

---

原因分析：

主要原因就是模型的key不一致，也就是层的名称不一致

比如我使用的是resnet模型，那么层名应该是conv1.weight而不是module.conv1.weight

---

解决方案：

方法1、 忽略层问题，强制导入模型的权重参数：

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

设置strict=False意味着在加载状态字典时，如果模型中有额外的键不匹配，将不会报错。这通常是安全的，但也可能导致某些问题，比如如果某些层的状态没有被正确加载。最好是确保模型结构与状态字典完全匹配，或者至少理解哪些部分可能不匹配。

虽然能够输出一些数值，但实际输出的分类output很可能是错误的

比如对于同一个样本同一个模型的每一次测试结果都不一样。

方法2、多GPU训练，多GPU测试

在测试环境中把单GPU定义为多GPU，定义关键字torch.nn.DataParallel()

import torch.nn as nn
import os

os.environ["CUDA_VISIBLE_DEVICES"] = "0, 1, 2, 3"

if __name__ == '__main__':
	# 此处为测试代码对接收参数的初始化
	parser = argparse.ArgumentParser()
	parser.add_argument('-gpu', action='store_true', default=True, help='use gpu or not')
	......
	......
	......

	# 加载网络
	net = get_network(args)
	
	# 以下为多GPU环境增加代码，以解决本博客中提到的报错问题
	net = net.cuda()
	net = nn.DataParallel(net, device_ids=[0, 1, 2, 3])
	
	# 以下为原始测试代码部分
	net.load_dict(torch.load(args.weights))
	net.eval()
	......
	......
	......
		

参考博客

方法3、修改模型key

conv1.weight与module.conv1.weight之间多了一个module

这是因为在加载多GPU训练的模型的时候，由于用DataParallel训练的模型数据并行方式的训练，key中会包含”module“关键字

去掉DataParallel 预训练模型中的module，修改如下：

model.load_state_dict({k.replace('module.',''):v for k,v in checkpoint['state_dict'].items()})

参考博客