如何将 torch.norm 转换为余弦距离

Question

我想把范数距离改成余弦距离，帮我把这个函数转换成余弦距离

def feat_prototype_distance(self, feat):
    N, C, H, W = feat.shape
    feat_proto_distance = -torch.ones((N, self.class_numbers, H, W)).to(feat.device)
    for i in range(self.class_numbers):
        feat_proto_distance[:, i, :, :] = torch.norm(self.objective_vectors[i].reshape(-1,1,1).expand(-1, H, W) - feat, 2, dim=1)
    return feat_proto_distance

这是使用带形状的范数距离的原始函数： self.objective_vectors[i].reshape(-1,1,1).expand(-1, H, W): torch.Size([256, 128, 224]) 壮举：torch.Size([8, 256, 128, 224]) with 8 is batch_size

Answer 1

你可以使用torch.nn.CosineSimilarity。

你的代码结构对我来说不是很清楚，但你可能会这样做：

def feat_prototype_distance(self, feat):
    distance_metric = torch.nn.CosineSimilarity(dim=1)
    N, C, H, W = feat.shape
    feat_proto_distance = -torch.ones((N, self.class_numbers, H, W)).to(feat.device)
    for i in range(self.class_numbers):
        feat_proto_distance[:, i, :, :] = distance_metric(self.objective_vectors[i].reshape(-1,1,1).expand(-1, H, W), feat)
    return feat_proto_distance