[pytorch] 图像识别之label smoothing (+mixup/cutmix)

def onehot_encoding(label, n_classes):
    return torch.zeros(label.size(0), n_classes).to(label.device).scatter_(
        1, label.view(-1, 1), 1)
def cross_entropy_loss(input, target, reduction):
    logp = F.log_softmax(input, dim=1)
    loss = torch.sum(-logp * target, dim=1)
    if reduction == 'none':
        return loss
    elif reduction == 'mean':
        return loss.mean()
    elif reduction == 'sum':
        return loss.sum()
    else:
        raise ValueError(
            '`reduction` must be one of \'none\', \'mean\', or \'sum\'.')
def label_smoothing_criterion(epsilon=0.1, reduction='mean'):
    def _label_smoothing_criterion(preds, targets):
        n_classes = preds.size(1)
        device = preds.device

        onehot = onehot_encoding(targets, n_cl