[Bug Fix] fix cross_entropy bug when 255 in label (#1499)

PaddlePaddle · Nov 5, 2021 · 11d7807 · 11d7807
1 parent 39addba
commit 11d7807
Showing 1 changed file with 2 additions and 4 deletions.
diff --git a/paddleseg/models/losses/cross_entropy_loss.py b/paddleseg/models/losses/cross_entropy_loss.py
@@ -48,11 +48,8 @@ def __init__(self,
         self.data_format = data_format
         if weight is not None:
             self.weight = paddle.to_tensor(weight, dtype='float32')
-            long_weight = weight + [0] * (256 - len(weight))
-            self.long_weight = paddle.to_tensor(long_weight, dtype='float32')
         else:
             self.weight = None
-            self.long_weight = None
 
     def forward(self, logit, label, semantic_weights=None):
         """
@@ -82,12 +79,13 @@ def forward(self, logit, label, semantic_weights=None):
         label = label.astype('int64')
 
         # In F.cross_entropy, the ignore_index is invalid, which needs to be fixed.
+        # When there is 255 in the label and paddle version <= 2.1.3, the cross_entropy OP will report an error, which is fixed in paddle develop version.
         loss = F.cross_entropy(
             logit,
             label,
             ignore_index=self.ignore_index,
             reduction='none',
-            weight=self.long_weight)
+            weight=self.weight)
 
         return self._post_process_loss(logit, label, semantic_weights, loss)