loss.py

import torch
import torch.nn as nn
import torch.nn.functional as F

def Entropy(input_):
    bs = input_.size(0)
    epsilon = 1e-5
    entropy = -input_ * torch.log(input_ + epsilon)
    entropy = torch.sum(entropy, dim=1)
    return entropy 

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, reduction=True):
        super(CrossEntropyLabelSmooth, self).__init__()
        self.num_classes = num_classes
        self.epsilon = epsilon
        self.use_gpu = use_gpu
        self.reduction = reduction
        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).cpu(), 1)
        if self.use_gpu: targets = targets.cuda()
        targets = (1 - self.epsilon) * targets + self.epsilon / self.num_classes
        loss = (- targets * log_probs).sum(dim=1)
        if self.reduction:
            return loss.mean()
        return loss

class SupConLoss(nn.Module):
    def __init__(self, temperature=0.07, contrast_mode='all',
                 base_temperature=0.07):
        super(SupConLoss, self).__init__()
        self.temperature = temperature
        self.contrast_mode = contrast_mode
        self.base_temperature = base_temperature

    def forward(self, anchor_feature, anc_labels, contrast_feature, con_labels):
        batch_size = anc_labels.shape[0]
        mask = torch.eq(anc_labels.unsqueeze(dim=1), con_labels.unsqueeze(dim=1).T).float().cuda()
        # compute logits
        anchor_dot_contrast = torch.div(
            torch.matmul(anchor_feature, contrast_feature.T),
            self.temperature)
        # for numerical stability
        logits_max, _ = torch.max(anchor_dot_contrast, dim=1, keepdim=True)
        logits = anchor_dot_contrast - logits_max.detach()

        # compute log_prob
        exp_logits = torch.exp(logits)
        log_prob = logits - torch.log(exp_logits.sum(1, keepdim=True))
        # compute mean of log-likelihood over positive
        mean_log_prob_pos = (mask * log_prob).sum(1) / (mask.sum(1) + 1e-5)

        # loss
        loss = - (self.temperature / self.base_temperature) * mean_log_prob_pos
        loss = loss.view(-1, batch_size).mean()

        return loss