revert for running

ppcls/arch/backbone/legendary_models/mobilenet_v1.py

@@ -165,7 +165,6 @@ class MobileNet(TheseusLayer):
             return_stages=return_stages)
 
     @AMP_forward_decorator
-    @clas_forward_decorator
     def forward(self, x):
         x = self.conv(x)
         x = self.blocks(x)

ppcls/engine/evaluation/classification.py

@@ -67,7 +67,7 @@ class ClassEval(object):
             if not self.config["Global"].get("use_multilabel", False):
                 batch[1] = batch[1].reshape([-1, 1]).astype("int64")
 
-            out = self.model(batch)
+            out = self.model(batch[0])
 
             # just for DistributedBatchSampler issue: repeat sampling
             current_samples = batch_size * paddle.distributed.get_world_size()

ppcls/engine/train/classification.py

@@ -41,6 +41,12 @@ class ClassTrainer(object):
         # gradient accumulation
         self.update_freq = self.config["Global"].get("update_freq", 1)
 
+        if "Head" in self.config["Arch"] or self.config["Arch"].get("is_rec",
+                                                                    False):
+            self.is_rec = True
+        else:
+            self.is_rec = False
+
         # TODO(gaotingquan): mv to build_model
         # build EMA model
         self.model_ema = self._build_ema_model()
@@ -197,7 +203,11 @@ class ClassTrainer(object):
                 batch[1] = batch[1].reshape([batch_size, -1])
             self.global_step += 1
 
-            out = self.model(batch)
+            if self.is_rec:
+                out = self.model(batch)
+            else:
+                out = self.model(batch[0])
+
             loss_dict = self.loss_func(out, batch[1])
             # TODO(gaotingquan): mv update_freq to loss and optimizer
             loss = loss_dict["loss"] / self.update_freq