def forward(self,
x,
params=None,
buffers=None,
bn_training=None,
do_training=None,
no_grad_hint_mask=None):
if params is None:
params = [p for p in self.parameters()]
if buffers is None:
buffers = [p for p in utils.get_buffers(self)]
if no_grad_hint_mask is None:
no_grad_hint_mask = '0' * len([m for m in self.modules()])
if bn_training is None:
bn_training = [self.training] * self.get_num_batchnorms()
if do_training is None:
do_training = self.training
out = self.feat_extractor(x,
params=params,
buffers=buffers,
bn_training=bn_training,
do_training=do_training,
no_grad_hint_mask=no_grad_hint_mask[1:])
return out.view(out.size(0), -1)
def forward(self,
x_input,
params=None,
buffers=None,
bn_training=None,
do_training=None,
no_grad_hint_mask=None):
if params is None:
params = [p for p in self.parameters()]
if buffers is None:
buffers = [p for p in get_buffers(self)]
if no_grad_hint_mask is None:
no_grad_hint_mask = '0' * len([m for m in self.modules()])
if bn_training is None:
bn_training = [self.training] * self.get_num_batchnorms()
if do_training is None:
do_training = self.training
zipped = list(
zip(self.children(), self.unmerge_parameters(params),
self.unmerge_buffers(buffers),
self.unmerge_bn_training(bn_training),
self.unmerge_no_grad_hint_mask(no_grad_hint_mask)))
x = x_input
for b, p, pb, bt, nghm in zipped:
x = b(x,
params=p,
buffers=pb,
bn_training=bt,
do_training=do_training,
no_grad_hint_mask=nghm)
return x
def forward(self,
x_input,
params=None,
buffers=None,
bn_training=None,
do_training=None,
no_grad_hint_mask=None):
if params is None:
params = [p for p in self.parameters()]
if buffers is None:
buffers = [p for p in get_buffers(self)]
if no_grad_hint_mask is None:
no_grad_hint_mask = '0' * len([m for m in self.modules()])
if bn_training is None:
bn_training = [self.training] * self.get_num_batchnorms()
if do_training is None:
do_training = self.training
assert len(no_grad_hint_mask) == 1
with (NoopContext if no_grad_hint_mask == '0' else torch.no_grad)():
x = F.conv2d(x_input,
self.get_hybrid_param(params, 'weight'),
bias=(self.get_hybrid_param(params, 'bias')
if self.has_hybrid_param('bias') else None),
stride=self.stride,
padding=self.padding,
dilation=self.dilation,
groups=self.groups)
assert len(bn_training) == 0
return x
def forward(self,
x_input,
params=None,
buffers=None,
bn_training=None,
do_training=None,
no_grad_hint_mask=None):
if params is None:
params = [p for p in self.parameters()]
if buffers is None:
buffers = [p for p in get_buffers(self)]
if no_grad_hint_mask is None:
no_grad_hint_mask = '0' * len([m for m in self.modules()])
if bn_training is None:
bn_training = [self.training] * self.get_num_batchnorms()
if do_training is None:
do_training = self.training
assert len(no_grad_hint_mask) == 1
with (NoopContext if no_grad_hint_mask == '0' else torch.no_grad)():
x = F.batch_norm(x_input,
self.get_hybrid_buffer(buffers, 'running_mean'),
self.get_hybrid_buffer(buffers, 'running_var'),
weight=self.get_hybrid_param(params, 'weight'),
bias=self.get_hybrid_param(params, 'bias'),
training=bn_training.pop(0),
momentum=self.momentum,
eps=self.eps)
assert len(bn_training) == 0
return x
def unmerge_buffers(self, buffers):
children = self.children()
if buffers is None:
return [None] * len([b for b in children])
buffers = [p for p in buffers]
nums_buffers = [len([p for p in get_buffers(b)]) for b in children]
assert (sum(nums_buffers) == len(buffers))
unmerged_buffers = []
for n in nums_buffers:
unmerged_buffers.append(buffers[:n])
buffers = buffers[n:]
return unmerged_buffers
def forward(self,
x,
params=None,
buffers=None,
bn_training=None,
do_training=None,
no_grad_hint_mask=None):
if params is None:
params = [p for p in self.parameters()]
if buffers is None:
buffers = [p for p in utils.get_buffers(self)]
if no_grad_hint_mask is None:
no_grad_hint_mask = '0' * len([m for m in self.modules()])
if bn_training is None:
bn_training = [self.training] * self.get_num_batchnorms()
if do_training is None:
do_training = self.training
zipped = list(
zip(self.children(), self.unmerge_parameters(params),
self.unmerge_buffers(buffers),
self.unmerge_bn_training(bn_training),
self.unmerge_no_grad_hint_mask(no_grad_hint_mask)))
assert len(zipped) == 2
b, p, pb, bt, nghm = zipped[0]
assert b is self.conv_block
conv_block_result = self.conv_block(x,
params=p,
buffers=pb,
bn_training=bt,
do_training=do_training,
no_grad_hint_mask=nghm)
b, p, pb, bt, nghm = zipped[1]
assert b is self.skip_layer
skip_layer_result = self.skip_layer(x,
params=p,
buffers=pb,
bn_training=bt,
do_training=do_training,
no_grad_hint_mask=nghm)
return skip_layer_result + conv_block_result