losses_intensity_weight.py

from __future__ import print_function

import torch
import torch.nn as nn

def compute_covariance(input_data):
    """
    Compute Covariance matrix of the input data
    """
    n = input_data.size(0)  # batch_size

    # Check if using gpu or cpu
    if input_data.is_cuda:
        device = torch.device('cuda')
    else:
        device = torch.device('cpu')

    id_row = torch.ones(n).resize(1, n).to(device=device)
    sum_column = torch.mm(id_row, input_data)
    mean_column = torch.div(sum_column, n)
    term_mul_2 = torch.mm(mean_column.t(), mean_column)
    d_t_d = torch.mm(input_data.t(), input_data)
    c = torch.add(d_t_d, (-1 * term_mul_2)) * 1 / (n - 1)

    return c


def covariance_loss(z1: torch.Tensor, z2: torch.Tensor) -> torch.Tensor:
    """Computes covariance loss given batch of projected features z1 from view 1 and
    projected features z2 from view 2.
    Args:
        z1 (torch.Tensor): NxD Tensor containing projected features from view 1.
        z2 (torch.Tensor): NxD Tensor containing projected features from view 2.
    Returns:
        torch.Tensor: covariance regularization loss.
    """

    N, D = z1.size()

    z1 = z1 - z1.mean(dim=0)
    z2 = z2 - z2.mean(dim=0)
    cov_z1 = (z1.T @ z1) / (N - 1)
    cov_z2 = (z2.T @ z2) / (N - 1)

    diag = torch.eye(D, device=z1.device)
    cov_loss = cov_z1[~diag.bool()].pow_(2).sum() / D + cov_z2[~diag.bool()].pow_(2).sum() / D
    return cov_loss

class DCCLoss(nn.Module):
    """Supervised Contrastive Learning: https://arxiv.org/pdf/2004.11362.pdf.
    It also supports the unsupervised contrastive loss in SimCLR"""
    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, features, labels=None, mask=None, image_weight=None):
        """Compute loss for model. If both `labels` and `mask` are None,
        it degenerates to SimCLR unsupervised loss:
        https://arxiv.org/pdf/2002.05709.pdf
        Args:
            features: hidden vector of shape [bsz, n_views, ...].
            labels: ground truth of shape [bsz].
            mask: contrastive mask of shape [bsz, bsz], mask_{i,j}=1 if sample j
                has the same class as sample i. Can be asymmetric.
        Returns:
            A loss scalar.
        """
        device = (torch.device('cuda')
                  if features.is_cuda
                  else torch.device('cpu'))

        if len(features.shape) < 3:
            raise ValueError('`features` needs to be [bsz, n_views, ...],'
                             'at least 3 dimensions are required')
        if len(features.shape) > 3:
            features = features.view(features.shape[0], features.shape[1], -1)

        batch_size = features.shape[0]
        if labels is not None and mask is not None:
            raise ValueError('Cannot define both `labels` and `mask`')
        elif labels is None and mask is None:
            mask = torch.eye(batch_size, dtype=torch.float32).to(device)
        elif labels is not None:
            labels = labels.contiguous().view(-1, 1)
            if labels.shape[0] != batch_size:
                raise ValueError('Num of labels does not match num of features')
            mask = torch.eq(labels, labels.T).float().to(device)
        else:
            mask = mask.float().to(device)

        contrast_count = features.shape[1]
        contrast_feature = torch.cat(torch.unbind(features, dim=1), dim=0)
        # tmp = compute_covariance(contrast_feature)
        
        convar_loss = covariance_loss(torch.unbind(features, dim=1)[0], torch.unbind(features, dim=1)[1])
        
        if self.contrast_mode == 'one':
            anchor_feature = features[:, 0]
            anchor_count = 1
        elif self.contrast_mode == 'all':
            anchor_feature = contrast_feature
            anchor_count = contrast_count
        else:
            raise ValueError('Unknown mode: {}'.format(self.contrast_mode))

        # compute logits
        anchor_dot_contrast = torch.div(
            torch.matmul(anchor_feature, contrast_feature.T),
            self.temperature)
        
        # self_covariance = compute_covariance(torch.matmul(anchor_feature, contrast_feature.T))
        # for numerical stability
        logits_max, _ = torch.max(anchor_dot_contrast, dim=1, keepdim=True)
        logits = anchor_dot_contrast - logits_max.detach()

        # tile mask
        mask = mask.repeat(anchor_count, contrast_count)
        # mask-out self-contrast cases
        logits_mask = torch.scatter(
            torch.ones_like(mask),
            1,
            torch.arange(batch_size * anchor_count).view(-1, 1).to(device),
            0
        )
        mask = mask * logits_mask 
        # * self_covariance

        # compute log_prob
        exp_logits = torch.exp(logits * image_weight) * logits_mask 
        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)

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

        return loss