Function classification_loss

loss.py:198–209

(prob_volume, depth_values, interval, depth_gt, mask, weight)

    return loss

def classification_loss(prob_volume, depth_values, interval, depth_gt, mask, weight):
    depth_gt_volume = depth_gt.unsqueeze(1).expand_as(depth_values)  # (b, d, h, w)

    gt_index_volume = (
            ((depth_values - interval / 2) <= depth_gt_volume).float() * ((depth_values + interval / 2) > depth_gt_volume).float())

    NEAR_0 = 1e-4  # Prevent overflow
    prob_volume = torch.where(prob_volume <= 0.0, torch.zeros_like(prob_volume) + NEAR_0, prob_volume)

    loss = -torch.sum(gt_index_volume * torch.log(prob_volume), dim=1)[mask].mean()
    loss = loss * weight
    return loss


def gfocal_loss(prob_volume, depth_values, interval, depth_gt, mask, weight, gamma, alpha):