multi-gpu ckpt filesize bug fix #253

train.py

@@ -287,7 +287,7 @@ def train(hyp):
         scheduler.step()
 
         # mAP
-        ema.update_attr(model)
+        ema.update_attr(model, include=['md', 'nc', 'hyp', 'names', 'stride'])
         final_epoch = epoch + 1 == epochs
         if not opt.notest or final_epoch:  # Calculate mAP
             results, maps, times = test.test(opt.data,

utils/torch_utils.py

@@ -173,22 +173,23 @@ def scale_img(img, ratio=1.0, same_shape=False):  # img(16,3,256,416), r=ratio
     return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447)  # value = imagenet mean
 
 
+def copy_attr(a, b, include=(), exclude=()):
+    # Copy attributes from b to a, options to only include [...] and to exclude [...]
+    for k, v in b.__dict__.items():
+        if (len(include) and k not in include) or k.startswith('_') or k in exclude:
+            continue
+        else:
+            setattr(a, k, v)
+
+
 class ModelEMA:
     """ Model Exponential Moving Average from https://github.com/rwightman/pytorch-image-models
     Keep a moving average of everything in the model state_dict (parameters and buffers).
     This is intended to allow functionality like
     https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
     A smoothed version of the weights is necessary for some training schemes to perform well.
-    E.g. Google's hyper-params for training MNASNet, MobileNet-V3, EfficientNet, etc that use
-    RMSprop with a short 2.4-3 epoch decay period and slow LR decay rate of .96-.99 requires EMA
-    smoothing of weights to match results. Pay attention to the decay constant you are using
-    relative to your update count per epoch.
-    To keep EMA from using GPU resources, set device='cpu'. This will save a bit of memory but
-    disable validation of the EMA weights. Validation will have to be done manually in a separate
-    process, or after the training stops converging.
     This class is sensitive where it is initialized in the sequence of model init,
     GPU assignment and distributed training wrappers.
-    I've tested with the sequence in my own train.py for torch.DataParallel, apex.DDP, and single-GPU.
     """
 
     def __init__(self, model, decay=0.9999, updates=0):
@@ -211,8 +212,6 @@ class ModelEMA:
                     v *= d
                     v += (1. - d) * msd[k].detach()
 
-    def update_attr(self, model):
+    def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
         # Update EMA attributes
-        for k, v in model.__dict__.items():
+        copy_attr(self.ema, model, include, exclude)
-            if not k.startswith('_') and k not in ["process_group", "reducer"]:
-                setattr(self.ema, k, v)