loss.py

import torch
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd.function import Function
from torch.autograd import Variable

class HcTripletLoss(nn.Module):
    """Hetero-center-triplet-loss-for-VT-Re-ID.
    
    Reference:
    Parameter Sharing Exploration and Hetero center triplet loss for VT Re-ID,TMM.
    Code imported from https://github.com/hijune6/Hetero-center-triplet-loss-for-VT-Re-ID/blob/main/loss.py.
    
    Args:
    - margin (float): margin for triplet.
    """
    
    def __init__(self, batch_size, margin=0.3):
        super(HcTripletLoss, self).__init__()
        self.margin = margin
        self.ranking_loss = nn.MarginRankingLoss(margin=margin)

    def forward(self, feats, labels):
        """
        Args:
        - inputs: feature matrix with shape (batch_size, feat_dim)
        - targets: ground truth labels with shape (num_classes)
        """
        label_uni = labels.unique()
        targets = torch.cat([label_uni,label_uni])
        label_num = len(label_uni)
        feat = feats.chunk(label_num*2, 0)
        center = []
        for i in range(label_num*2):
            center.append(torch.mean(feat[i], dim=0, keepdim=True))
        inputs = torch.cat(center)

        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)
        loss = self.ranking_loss(dist_an, dist_ap, y)
        
        # compute accuracy
        correct = torch.ge(dist_an, dist_ap).sum().item()
        return loss, correct


class OriTripletLoss(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.
    Code imported from https://github.com/Cysu/open-reid/blob/master/reid/loss/triplet.py.
    
    Args:
    - margin (float): margin for triplet.
    """
    
    def __init__(self, batch_size, margin=0.3):
        super(OriTripletLoss, self).__init__()
        self.margin = margin
        self.ranking_loss = nn.MarginRankingLoss(margin=margin)

    def forward(self, inputs, targets):
        """
        Args:
        - inputs: feature matrix with shape (batch_size, feat_dim)
        - targets: 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)
        loss = self.ranking_loss(dist_an, dist_ap, y)
        
        # compute accuracy
        correct = torch.ge(dist_an, dist_ap).sum().item()
        return loss, correct

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