loss.py

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable


class FocalLoss(nn.Module):
    r"""
        This criterion is a implemenation of Focal Loss, which is proposed in
        Focal Loss for Dense Object Detection.
            Loss(x, class) = - \alpha (1-softmax(x)[class])^gamma \log(softmax(x)[class])
        The losses are averaged across observations for each minibatch.
        Args:
            alpha(1D Tensor, Variable) : the scalar factor for this criterion
            gamma(float, double) : gamma > 0; reduces the relative loss for well-classified examples (p > .5),
                                   putting more focus on hard, misclassified examples
            size_average(bool): By default, the losses are averaged over observations for each minibatch.
                                However, if the field size_average is set to False, the losses are
                                instead summed for each minibatch.
    """

    def __init__(self, class_num, alpha=None, gamma=2, size_average=True):
        super(FocalLoss, self).__init__()
        if alpha is None:
            self.alpha = nn.Parameter(torch.ones(class_num, 1))
        else:
            if isinstance(alpha, nn.Parameter):
                self.alpha = alpha
            else:
                self.alpha = nn.Parameter(torch.tensor(alpha))
        self.gamma = gamma
        self.class_num = class_num
        self.size_average = size_average

    def forward(self, inputs, targets):
        N = inputs.size(0)
        C = inputs.size(1)
        # dim = 1 是自己加的
        P = F.softmax(inputs, dim=1)

        class_mask = inputs.data.new(N, C).fill_(0)
        class_mask = torch.tensor(class_mask)
        ids = targets.view(-1, 1)
        class_mask.scatter_(1, ids.data, 1.)
        # print(class_mask)

        # self.alpha = self.alpha.to(inputs.device)
        alpha = self.alpha[ids.data.view(-1)]

        probs = (P * class_mask).sum(1).view(-1, 1)

        log_p = probs.log()
        # print('probs size= {}'.format(probs.size()))
        # print(probs)

        batch_loss = -alpha * (torch.pow((1 - probs), self.gamma)) * log_p
        # print('-----bacth_loss------')
        # print(batch_loss)

        if self.size_average:
            loss = batch_loss.mean()
        else:
            loss = batch_loss.sum()
        return loss


if __name__ == '__main__':
    net = FocalLoss(3)
    net.train()
    import torch.optim as optim

    optimizer = optim.Adam(net.parameters(), lr=0.01)
    for i in range(1000):
        optimizer.zero_grad()
        input_ids = torch.rand(4, 3)
        labels = torch.randint(0, 3, (1, 4))
        loss = abs(net(input_ids, labels))
        print(loss.item(), net.alpha)
        loss.backward()
        optimizer.step()