关于 dense relation distill module

[Saralyliu]

def forward(self,features,attentions):
    features = list(features)
    if isinstance(attentions,dict):
        for i in range(len(attentions)):
            if i==0:
                atten = attentions[i].unsqueeze(0)
            else:
                atten = torch.cat((atten,attentions[i].unsqueeze(0)),dim=0)
        attentions = atten.cuda()
    output = []
    h , w = attentions.shape[2:]
    ncls = attentions.shape[0]       
    key_t = self.key_t(attentions)   
    val_t = self.value_t(attentions) 
    for idx in range(len(features)):

我想知道谁是support ，谁是query？按代码的意思，是attentions是support，features是query是这样吗？
attentions和features的维度分别是多少？
期待您的回复，谢谢！