fastreid.modeling.losses.cross_entroy_loss Namespace Reference

Functions
	log_accuracy (pred_class_logits, gt_classes, topk=(1,))
	cross_entropy_loss (pred_class_logits, gt_classes, eps, alpha=0.2)

Detailed Description

@author:  l1aoxingyu
@contact: sherlockliao01@gmail.com

Function Documentation

cross_entropy_loss()

fastreid.modeling.losses.cross_entroy_loss.cross_entropy_loss	(		pred_class_logits,
			gt_classes,
			eps,
			alpha = 0.2 )

Definition at line 31 of file cross_entroy_loss.py.

def cross_entropy_loss(pred_class_logits, gt_classes, eps, alpha=0.2):
    num_classes = pred_class_logits.size(1)
 
    if eps >= 0:
        smooth_param = eps
    else:
        # Adaptive label smooth regularization
        soft_label = F.softmax(pred_class_logits, dim=1)
        smooth_param = alpha * soft_label[torch.arange(soft_label.size(0)), gt_classes].unsqueeze(1)
 
    log_probs = F.log_softmax(pred_class_logits, dim=1)
    with torch.no_grad():
        targets = torch.ones_like(log_probs)
        targets *= smooth_param / (num_classes - 1)
        targets.scatter_(1, gt_classes.data.unsqueeze(1), (1 - smooth_param))
 
    loss = (-targets * log_probs).sum(dim=1)
 
    """
    # confidence penalty
    conf_penalty = 0.3
    probs = F.softmax(pred_class_logits, dim=1)
    entropy = torch.sum(-probs * log_probs, dim=1)
    loss = torch.clamp_min(loss - conf_penalty * entropy, min=0.)
    """
 
    with torch.no_grad():
        non_zero_cnt = max(loss.nonzero(as_tuple=False).size(0), 1)
 
    loss = loss.sum() / non_zero_cnt
 
    return loss

log_accuracy()

fastreid.modeling.losses.cross_entroy_loss.log_accuracy	(		pred_class_logits,
			gt_classes,
			topk = (1,) )

Log the accuracy metrics to EventStorage.

Definition at line 12 of file cross_entroy_loss.py.

def log_accuracy(pred_class_logits, gt_classes, topk=(1,)):
    """
    Log the accuracy metrics to EventStorage.
    """
    bsz = pred_class_logits.size(0)
    maxk = max(topk)
    _, pred_class = pred_class_logits.topk(maxk, 1, True, True)
    pred_class = pred_class.t()
    correct = pred_class.eq(gt_classes.view(1, -1).expand_as(pred_class))
 
    ret = []
    for k in topk:
        correct_k = correct[:k].view(-1).float().sum(dim=0, keepdim=True)
        ret.append(correct_k.mul_(1. / bsz))
 
    storage = get_event_storage()
    storage.put_scalar("cls_accuracy", ret[0])