class Loader(object):
def __init__(self, args):
self.args = args
if self.args.dataset == 'cityscapes':
self.nclass = 19
elif self.args.dataset in ['2d','3d']:
self.nclass = 4
# Resuming checkpoint
self.best_pred = 0.0
assert args.resume is not None, RuntimeError("No model to decode in resume path: '{:}'".format(args.resume))
assert os.path.isfile(args.resume), RuntimeError("=> no checkpoint found at '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
self._alphas = checkpoint['state_dict']['alphas']
self._betas = checkpoint['state_dict']['betas']
self.decoder = Decoder(alphas=self._alphas, betas=self._betas, steps=5)
def retreive_alphas_betas(self):
return self._alphas, self._betas
def decode_architecture(self):
paths, paths_space = self.decoder.viterbi_decode()
return paths, paths_space
def decode_cell(self):
genotype = self.decoder.genotype_decode()
return genotype
class Loader(object):
def __init__(self, args):
self.args = args
if self.args.dataset == 'cityscapes':
self.nclass = 19
self.model = AutoDeeplab(num_classes=self.nclass,
num_layers=12,
filter_multiplier=self.args.filter_multiplier,
block_multiplier=args.block_multiplier,
step=args.step)
# Using cuda
if args.cuda:
if (torch.cuda.device_count() > 1 or args.load_parallel):
self.model = torch.nn.DataParallel(self.model.cuda())
patch_replication_callback(self.model)
self.model = self.model.cuda()
print('cuda finished')
# Resuming checkpoint
self.best_pred = 0.0
if args.resume is not None:
if not os.path.isfile(args.resume):
raise RuntimeError("=> no checkpoint found at '{}'".format(
args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
# if the weights are wrapped in module object we have to clean it
if args.clean_module:
self.model.load_state_dict(checkpoint['state_dict'])
state_dict = checkpoint['state_dict']
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[7:] # remove 'module.' of dataparallel
new_state_dict[name] = v
self.model.load_state_dict(new_state_dict)
else:
if (torch.cuda.device_count() > 1 or args.load_parallel):
self.model.module.load_state_dict(checkpoint['state_dict'])
else:
self.model.load_state_dict(checkpoint['state_dict'])
self.decoder = Decoder(self.model.alphas, self.model.bottom_betas,
self.model.betas8, self.model.betas16,
self.model.top_betas, args.block_multiplier,
args.step)
def retreive_alphas_betas(self):
return self.model.alphas, self.model.bottom_betas, self.model.betas8, self.model.betas16, self.model.top_betas
def decode_architecture(self):
paths, paths_space = self.decoder.viterbi_decode()
return paths, paths_space
def decode_cell(self):
genotype = self.decoder.genotype_decode()
return genotype
def __init__(self, args):
self.args = args
if self.args.dataset == 'cityscapes':
self.nclass = 19
elif self.args.dataset in ['2d','3d']:
self.nclass = 4
# Resuming checkpoint
self.best_pred = 0.0
assert args.resume is not None, RuntimeError("No model to decode in resume path: '{:}'".format(args.resume))
assert os.path.isfile(args.resume), RuntimeError("=> no checkpoint found at '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
self._alphas = checkpoint['state_dict']['alphas']
self._betas = checkpoint['state_dict']['betas']
self.decoder = Decoder(alphas=self._alphas, betas=self._betas, steps=5)
def genotype(self):
decoder = Decoder(self.alphas_d, self.alphas_c, self.betas, self._step)
return decoder.genotype_decode()
def decode_dfs(self):
decoder = Decoder(self.alphas_d, self.alphas_c, self.betas, 5)
return decoder.dfs_decode()
def decode_viterbi(self):
decoder = Decoder(self.alphas_d, self.alphas_c, self.betas, 5)
return decoder.viterbi_decode()
def genotype(self):
decoder = Decoder(self.alphas_cell, self._block_multiplier, self._step)
return decoder.genotype_decode()
def decode_dfs(self):
decoder = Decoder(self.bottom_betas, self.betas8, self.betas16, self.top_betas)
return decoder.dfs_decode()
def decode_viterbi(self):
decoder = Decoder(self.bottom_betas, self.betas8, self.betas16, self.top_betas)
return decoder.viterbi_decode()