EMA bug fix #279

train.py

@@ -294,7 +294,7 @@ def train(hyp):
                                              batch_size=batch_size,
                                              imgsz=imgsz_test,
                                              save_json=final_epoch and opt.data.endswith(os.sep + 'coco.yaml'),
-                                             model=ema.ema.module if hasattr(model, 'module') else ema.ema,
+                                             model=ema.ema,
                                              single_cls=opt.single_cls,
                                              dataloader=testloader)
 
@@ -324,7 +324,7 @@ def train(hyp):
                 ckpt = {'epoch': epoch,
                         'best_fitness': best_fitness,
                         'training_results': f.read(),
-                        'model': ema.ema.module if hasattr(model, 'module') else ema.ema,
+                        'model': ema.ema,
                         'optimizer': None if final_epoch else optimizer.state_dict()}
 
             # Save last, best and delete

utils/torch_utils.py

@@ -175,8 +175,8 @@ class ModelEMA:
     """
 
     def __init__(self, model, decay=0.9999, device=''):
-        # make a copy of the model for accumulating moving average of weights
-        self.ema = deepcopy(model)
+        # Create EMA
+        self.ema = deepcopy(model.module if is_parallel(model) else model).half()  # FP16 EMA
         self.ema.eval()
         self.updates = 0  # number of EMA updates
         self.decay = lambda x: decay * (1 - math.exp(-x / 2000))  # decay exponential ramp (to help early epochs)
@@ -187,22 +187,19 @@ class ModelEMA:
             p.requires_grad_(False)
 
     def update(self, model):
-        self.updates += 1
-        d = self.decay(self.updates)
+        # Update EMA parameters
         with torch.no_grad():
-            if is_parallel(model):
-                msd, esd = model.module.state_dict(), self.ema.module.state_dict()
-            else:
-                msd, esd = model.state_dict(), self.ema.state_dict()
+            self.updates += 1
+            d = self.decay(self.updates)
 
-            for k, v in esd.items():
+            msd = model.module.state_dict() if is_parallel(model) else model.state_dict()  # model state_dict
+            for k, v in self.ema.state_dict().items():
                 if v.dtype.is_floating_point:
                     v *= d
                     v += (1. - d) * msd[k].detach()
 
     def update_attr(self, model):
-        # Update class attributes
-        ema = self.ema.module if is_parallel(model) else self.ema
+        # Update EMA attributes
         for k, v in model.__dict__.items():
             if not k.startswith('_') and k != 'module':
-                setattr(ema, k, v)
+                setattr(self.ema, k, v)