def load(self, path):
"""
Loads saved weights into the model
"""
with path.open('rb') as f:
weights = torch.load(f)
load_model_from_dict(self, weights)
return self
def load(self, path):
"""
Loads saved weights into the model
"""
with path.open('rb') as f:
weights = torch.load(f)
load_model_from_dict(self.align, weights)
# model_params = dict(self.align.named_parameters())
# model_keys = sorted(model_params.keys())
# print(model_keys)
return self
def load(archive_path=None,
height=5,
dim=1536,
skips=0,
topskips=0,
k=7,
cuda=True,
num_targets=1):
"""
Builds and load a model with the specified architecture from
an archive.
Params:
height: the number of layers in the pyramid (including
bottom layer (number of downsamples = height - 1)
dim: the size of the full resolution images used as input
skips: the number of residual fields (from the bottom of the
pyramid) to skip
cuda: whether or not to move the model to the GPU
"""
assert archive_path is not None, "Must provide an archive"
model = PyramidTransformer(size=height,
dim=dim,
k=k,
skip=skips,
topskips=topskips,
num_targets=num_targets)
if cuda:
model = model.cuda()
for p in model.parameters():
p.requires_grad = False
model.train(False)
print('Loading model state from {}...'.format(archive_path))
state_dict = torch.load(archive_path)
load_model_from_dict(model, state_dict)
print('Successfully loaded model state.')
return model