triplet_loss.py

# encoding: utf-8
import torch
from typing import Tuple
from torch import nn, Tensor
import torch.nn.functional as F

def normalize(x, axis=-1):
    """Normalizing to unit length along the specified dimension.
    Args:
      x: pytorch Variable
    Returns:
      x: pytorch Variable, same shape as input
    """
    x = 1. * x / (torch.norm(x, 2, axis, keepdim=True).expand_as(x) + 1e-12)
    return x

def euclidean_dist(x, y):
    """
    计算距离矩阵函数
    dist=a-b=sqrt(a^2+b^2-2ab)
    Args:
      x: pytorch Variable, with shape [m, d]
      y: pytorch Variable, with shape [n, d]
    Returns:
      dist: pytorch Variable, with shape [m, n]
    """
    m, n = x.size(0), y.size(0)
    xx = torch.pow(x, 2).sum(1, keepdim=True).expand(m, n)
    yy = torch.pow(y, 2).sum(1, keepdim=True).expand(n, m).t()
    dist = xx + yy
    dist.addmm_(1, -2, x, y.t())
    dist = dist.clamp(min=1e-12).sqrt()  # for numerical stability
    return dist

def hard_example_mining(dist_mat, labels, return_inds=False):
    """For each anchor, find the hardest positive and negative sample.
    Args:
      dist_mat: pytorch Variable, pair wise distance between samples, shape [N, N]
      labels: pytorch LongTensor, with shape [N]
      return_inds: whether to return the indices. Save time if `False`(?)
    Returns:
      dist_ap: pytorch Variable, distance(anchor, positive); shape [N]
      dist_an: pytorch Variable, distance(anchor, negative); shape [N]
      p_inds: pytorch LongTensor, with shape [N];
        indices of selected hard positive samples; 0 <= p_inds[i] <= N - 1
      n_inds: pytorch LongTensor, with shape [N];
        indices of selected hard negative samples; 0 <= n_inds[i] <= N - 1
    NOTE: Only consider the case in which all labels have same num of samples,
      thus we can cope with all anchors in parallel.
    """

    assert len(dist_mat.size()) == 2
    assert dist_mat.size(0) == dist_mat.size(1)
    N = dist_mat.size(0)

    # shape [N, N]
    is_pos = labels.expand(N, N).eq(labels.expand(N, N).t())
    is_neg = labels.expand(N, N).ne(labels.expand(N, N).t())

    # `dist_ap` means distance(anchor, positive)
    # both `dist_ap` and `relative_p_inds` with shape [N, 1]
    dist_ap, relative_p_inds = torch.max(
        dist_mat[is_pos].contiguous().view(N, -1), 1, keepdim=True)
    # `dist_an` means distance(anchor, negative)
    # both `dist_an` and `relative_n_inds` with shape [N, 1]
    dist_an, relative_n_inds = torch.min(
        dist_mat[is_neg].contiguous().view(N, -1), 1, keepdim=True)
    # shape [N]
    dist_ap = dist_ap.squeeze(1)
    dist_an = dist_an.squeeze(1)

    if return_inds:
        # shape [N, N]
        ind = (labels.new().resize_as_(labels)
               .copy_(torch.arange(0, N).long())
               .unsqueeze(0).expand(N, N))
        # shape [N, 1]
        p_inds = torch.gather(
            ind[is_pos].contiguous().view(N, -1), 1, relative_p_inds.data)
        n_inds = torch.gather(
            ind[is_neg].contiguous().view(N, -1), 1, relative_n_inds.data)
        # shape [N]
        p_inds = p_inds.squeeze(1)
        n_inds = n_inds.squeeze(1)
        return dist_ap, dist_an, p_inds, n_inds

    return dist_ap, dist_an


class TripletLoss(object):
    """Modified from Tong Xiao's open-reid (https://github.com/Cysu/open-reid).
    Related Triplet Loss theory can be found in paper 'In Defense of the Triplet
    Loss for Person Re-Identification'."""

    def __init__(self, margin=None):
        self.margin = margin
        if margin is not None:
            self.ranking_loss = nn.MarginRankingLoss(margin=margin)
        else:
            self.ranking_loss = nn.SoftMarginLoss()

    def __call__(self, global_feat, labels, normalize_feature=False):
        if normalize_feature:
            global_feat = normalize(global_feat, axis=-1)
        dist_mat = euclidean_dist(global_feat, global_feat)
        dist_ap, dist_an = hard_example_mining(
            dist_mat, labels)
        y = dist_an.new().resize_as_(dist_an).fill_(1)
        if self.margin is not None:
            loss = self.ranking_loss(dist_an, dist_ap, y)
        else:
            loss = self.ranking_loss(dist_an - dist_ap, y)
        return loss, dist_ap, dist_an

class CrossEntropyLabelSmooth(nn.Module):
    """Cross entropy loss with label smoothing regularizer.

    Reference:
    Szegedy et al. Rethinking the Inception Architecture for Computer Vision. CVPR 2016.
    Equation: y = (1 - epsilon) * y + epsilon / K.

    Args:
        num_classes (int): number of classes.
        epsilon (float): weight.
    """
    def __init__(self, num_classes, epsilon=0.1, use_gpu=True):
        super(CrossEntropyLabelSmooth, self).__init__()
        self.num_classes = num_classes
        self.epsilon = epsilon
        self.use_gpu = use_gpu
        self.logsoftmax = nn.LogSoftmax(dim=1)

    def forward(self, inputs, targets):
        """
        Args:
            inputs: prediction matrix (before softmax) with shape (batch_size, num_classes)
            targets: ground truth labels with shape (num_classes)
        """
        log_probs = self.logsoftmax(inputs)
        targets = torch.zeros(log_probs.size()).scatter_(1, targets.unsqueeze(1).data.cpu(), 1)
        if self.use_gpu: targets = targets.cuda()
        targets = (1 - self.epsilon) * targets + self.epsilon / self.num_classes
        loss = (- targets * log_probs).mean(0).sum()
        return loss
    
############################################    ################################################

def get_world_size() -> int:
    if not torch.distributed.is_available():
        return 1
    if not torch.distributed.is_initialized():
        return 1
    return torch.distributed.get_world_size()

class CircleLoss(object):
    def __init__(self):
        self._scale = 1

        self.m = 0.25
        self.s = 128

    def __call__(self, embedding, targets):
        embedding = nn.functional.normalize(embedding, dim=1)

        if get_world_size() > 1:
            all_embedding = concat_all_gather(embedding)
            all_targets = concat_all_gather(targets)
        else:
            all_embedding = embedding
            all_targets = targets

        dist_mat = torch.matmul(embedding, all_embedding.t())

        N, M = dist_mat.size()
        is_pos = targets.view(N, 1).expand(N, M).eq(all_targets.view(M, 1).expand(M, N).t())
        is_neg = targets.view(N, 1).expand(N, M).ne(all_targets.view(M, 1).expand(M, N).t())

        s_p = dist_mat[is_pos].contiguous().view(N, -1)
        s_n = dist_mat[is_neg].contiguous().view(N, -1)

        alpha_p = F.relu(-s_p.detach() + 1 + self.m)
        alpha_n = F.relu(s_n.detach() + self.m)
        delta_p = 1 - self.m
        delta_n = self.m

        logit_p = - self.s * alpha_p * (s_p - delta_p)
        logit_n = self.s * alpha_n * (s_n - delta_n)

        loss = nn.functional.softplus(torch.logsumexp(logit_p, dim=1) + torch.logsumexp(logit_n, dim=1)).mean()

        return loss * self._scale
#####################################################################################################


class HardTripletLoss(nn.Module):
    """Triplet loss with hard positive/negative mining.
    
    Reference:
        Hermans et al. In Defense of the Triplet Loss for Person Re-Identification. arXiv:1703.07737.
    
    Imported from `<https://github.com/Cysu/open-reid/blob/master/reid/loss/triplet.py>`_.
    
    Args:
        margin (float, optional): margin for triplet. Default is 0.3.
    """

    def __init__(self, margin=0.3):
        super(HardTripletLoss, self).__init__()
        self.margin = margin
        self.ranking_loss = nn.MarginRankingLoss(margin=margin)

    def forward(self, inputs, targets):
        """
        Args:
            inputs (torch.Tensor): feature matrix with shape (batch_size, feat_dim).
            targets (torch.LongTensor): ground truth labels with shape (num_classes).
        """
        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().unsqueeze(0))
            dist_an.append(dist[i][mask[i] == 0].min().unsqueeze(0))
        dist_ap = torch.cat(dist_ap)
        dist_an = torch.cat(dist_an)

        # Compute ranking hinge loss
        y = torch.ones_like(dist_an)
        return self.ranking_loss(dist_an, dist_ap, y)
    
##############################################################################################


class CenterLoss(nn.Module):
    """Center loss.
    Reference:
    Wen et al. A Discriminative Feature Learning Approach for Deep Face Recognition. ECCV 2016.
    Args:
        num_classes (int): number of classes.
        feat_dim (int): feature dimension.
    """

    def __init__(self, num_classes=751, feat_dim=6144, use_gpu=True):
        super(CenterLoss, self).__init__()
        self.num_classes = num_classes
        self.feat_dim = feat_dim
        self.use_gpu = use_gpu

        if self.use_gpu:
            self.centers = nn.Parameter(torch.randn(self.num_classes, self.feat_dim).cuda())
        else:
            self.centers = nn.Parameter(torch.randn(self.num_classes, self.feat_dim))

    def forward(self, x, labels):
        """
        Args:
            x: feature matrix with shape (batch_size, feat_dim).
            labels: ground truth labels with shape (num_classes).
        """
        assert x.size(0) == labels.size(0), "features.size(0) is not equal to labels.size(0)"

        batch_size = x.size(0)
        distmat = torch.pow(x, 2).sum(dim=1, keepdim=True).expand(batch_size, self.num_classes) + \
                  torch.pow(self.centers, 2).sum(dim=1, keepdim=True).expand(self.num_classes, batch_size).t()
        distmat.addmm_(1, -2, x, self.centers.t())

        classes = torch.arange(self.num_classes).long()
        if self.use_gpu: classes = classes.cuda()
        labels = labels.unsqueeze(1).expand(batch_size, self.num_classes)
        mask = labels.eq(classes.expand(batch_size, self.num_classes))

        dist = distmat * mask.float()
        loss = dist.clamp(min=1e-12, max=1e+12).sum() / batch_size
        #dist = []
        #for i in range(batch_size):
        #    value = distmat[i][mask[i]]
        #    value = value.clamp(min=1e-12, max=1e+12)  # for numerical stability
        #    dist.append(value)
        #dist = torch.cat(dist)
        #loss = dist.mean()
        return loss