解决Pytorch修改预训练模型时遇到key不匹配的情况

#加载后保存（未修改网络）
base_weights = torch.load(args.save_folder + args.basenet)
ssd_net.vgg.load_state_dict(base_weights) 
torch.save(ssd_net.state_dict(), args.save_folder + 'ssd_base' + '.pth')

# 将新保存的网络代替之前的预训练模型
    ssd_net = build_ssd('train', cfg['min_dim'], cfg['num_classes'])
    net = ssd_net
    ...
    if args.resume:
        ...
    else:
        base_weights = torch.load(args.save_folder + args.basenet)
        #args.basenet为ssd_base.pth
        print('Loading base network...')
        ssd_net.vgg.load_state_dict(base_weights) 

from collections import OrderedDict   #导入此模块
base_weights = torch.load(args.save_folder + args.basenet)
print('Loading base network...')
new_state_dict = **OrderedDict()**  
for k, v in base_weights.items():
    name = k[4:]   # remove `vgg.`，即只取vgg.0.weights的后面几位
    new_state_dict[name] = v 
    ssd_net.vgg.load_state_dict(new_state_dict) 

   def load_specific_param(self, state_dict, param_name, model_path):
        param_dict = torch.load(model_path)
        for i in state_dict:
            key = param_name + '.' + i
            state_dict[i].copy_(param_dict[key])
        del param_dict

   def load_specific_param(self, state_dict, param_name, model_path):
        param_dict = torch.load(model_path)
        param_dict = {k: v for k, v in param_dict.items() if 'num_batches_tracked' not in k}
        for i in state_dict:
            key = param_name + '.' + i
            if 'num_batches_tracked' in key:
                continue
            state_dict[i].copy_(param_dict[key])
        del param_dict