• Search by Module
• Search by Words
• Search Projects

• Java
• Python
• JavaScript
• TypeScript
• C++
• Scala
• Blog

• Exploit-Unknown-Gradually-master
  • run.py
  • examples
  • LICENSE
  • run.sh
  • data
    • .gitkeep
  • README.md
  • logs
    • .gitkeep
  • reid
    • datasets
      • mars.py
      • dukemtmc_videoReID.py
      • __init__.py
    • utils
      • meters.py
      • osutils.py
      • serialization.py
      • data
        • sampler.py
        • preprocessor.py
        • __init__.py
        • dataset.py
        • transforms.py
      • __init__.py
      • logging.py
    • evaluation_metrics
      • ranking.py
      • classification.py
      • __init__.py
    • evaluators.py
    • trainers.py
    • models
      • resnet.py
      • end2end.py
      • __init__.py
    • dist_metric.py
    • feature_extraction
      • database.py
      • cnn.py
      • __init__.py
    • __init__.py
    • metric_learning
      • kissme.py
      • euclidean.py
      • __init__.py
    • loss
      • tri_clu_loss.py
      • triplet.py
      • __init__.py
      • oim.py
    • eug.py
  • .gitignore

from __future__ import absolute_import

import torch
from torch import nn
from torch.autograd import Variable


class TripletLoss(nn.Module):
    def __init__(self, margin=0):
        super(TripletLoss, self).__init__()
        self.margin = margin
        self.ranking_loss = nn.MarginRankingLoss(margin=margin)

    def forward(self, inputs, targets):
        n = inputs.size(0)
        # Compute pairwise distance, replace by the official when merged
        dist = torch.pow(inputs, 2).sum(dim=1, keepdim=True).expand(n, n)
        dist = dist + dist.t()
        dist.addmm_(1, -2, inputs, inputs.t())
        dist = dist.clamp(min=1e-12).sqrt()  # for numerical stability
        # For each anchor, find the hardest positive and negative
        mask = targets.expand(n, n).eq(targets.expand(n, n).t())
        dist_ap, dist_an = [], []
        for i in range(n):
            dist_ap.append(dist[i][mask[i]].max())
            dist_an.append(dist[i][mask[i] == 0].min())
        dist_ap = torch.cat(dist_ap)
        dist_an = torch.cat(dist_an)
        # Compute ranking hinge loss
        y = dist_an.data.new()
        y.resize_as_(dist_an.data)
        y.fill_(1)
        y = Variable(y)
        loss = self.ranking_loss(dist_an, dist_ap, y)
        prec = (dist_an.data > dist_ap.data).sum() * 1. / y.size(0)
        return loss, prec