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.
This class is sensitive where it is initialized in the sequence of model init,
GPU assignment and distributed training wrappers.
"""
def __init__(self, model, decay=0.9999, updates=0):
# Create EMA
self.ema = deepcopy(model.module if is_parallel(model) else model).eval() # FP32 EMA
# if next(model.parameters()).device.type != 'cpu':
# self.ema.half() # FP16 EMA
self.updates = updates # number of EMA updates
self.decay = lambda x: decay * (1 - math.exp(-x / 2000)) # decay exponential ramp (to help early epochs)
for p in self.ema.parameters():
p.requires_grad_(False)
def update(self, model):
# Update EMA parameters
with torch.no_grad():
self.updates += 1
d = self.decay(self.updates)
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, include=(), exclude=('process_group', 'reducer')):
# Update EMA attributes
copy_attr(self.ema, model, include, exclude)
滑动平均的公式为
m
v
t
=
d
e
c
a
y
∗
m
v
t
−
1
+
(
1
−
d
e
c
a
y
)
∗
v
a
r
i
a
b
l
e
mv_{t} = decay*mv_{t-1}+(1-decay)*variable
mvt=decay∗mvt−1+(1−decay)∗variable
对应上述代码中的
v *= d
v += (1 - d) * msd[k].detach()
具体细节再次不赘述，ema是用来在test或val时才使用，用来更新参数。
