More appropriate/correct loss name

Author: rwightman

loss/__init__.py

@@ -1 +1 @@
-from loss.cross_entropy import LabelSmoothingCrossEntropy, SparseLabelCrossEntropy
+from loss.cross_entropy import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy

loss/cross_entropy.py

@@ -26,10 +26,10 @@ def forward(self, x, target):
         return loss.mean()
 
 
-class SparseLabelCrossEntropy(nn.Module):
+class SoftTargetCrossEntropy(nn.Module):
 
     def __init__(self):
-        super(SparseLabelCrossEntropy, self).__init__()
+        super(SoftTargetCrossEntropy, self).__init__()
 
     def forward(self, x, target):
         loss = torch.sum(-target * F.log_softmax(x, dim=-1), dim=-1)

train.py

@@ -13,7 +13,7 @@
 from data import Dataset, create_loader, resolve_data_config, FastCollateMixup, mixup_target
 from models import create_model, resume_checkpoint
 from utils import *
-from loss import LabelSmoothingCrossEntropy, SparseLabelCrossEntropy
+from loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
 from optim import create_optimizer
 from scheduler import create_scheduler
 
@@ -261,7 +261,7 @@ def main():
 
     if args.mixup > 0.:
         # smoothing is handled with mixup label transform
-        train_loss_fn = SparseLabelCrossEntropy().cuda()
+        train_loss_fn = SoftTargetCrossEntropy().cuda()
         validate_loss_fn = nn.CrossEntropyLoss().cuda()
     elif args.smoothing:
         train_loss_fn = LabelSmoothingCrossEntropy(smoothing=args.smoothing).cuda()