train_triplet.py 文件源码

def forward(self, anchor, positive, negative):
        d_p = self.pdist.forward(anchor, positive)
        d_n = self.pdist.forward(anchor, negative)

        dist_hinge = torch.clamp(self.margin + d_p - d_n, min=0.0)
        loss = torch.mean(dist_hinge)
        return loss