def update_params(self):
r"""
Load model parameters
"""
self._make_convs()
self._make_nonlocal()
self._initialize_conv()
self._initialize_nonlocal()
super().update_params()
if self._hyper_params["pretrain_model_path"] != "":
model_path = self._hyper_params["pretrain_model_path"]
try:
state_dict = torch.load(model_path,
map_location=torch.device("gpu"))
except:
state_dict = torch.load(model_path,
map_location=torch.device("cpu"))
if "model_state_dict" in state_dict:
state_dict = state_dict["model_state_dict"]
try:
self.load_state_dict(state_dict, strict=True)
except:
self.load_state_dict(state_dict, strict=False)
logger.info("Pretrained weights loaded from {}".format(model_path))
logger.info("Check md5sum of Pretrained weights: %s" %
md5sum(model_path))
def update_params(self):
model_file = self._hyper_params["pretrain_model_path"]
if model_file != "":
state_dict = torch.load(model_file,
map_location=torch.device("cpu"))
self.load_state_dict(state_dict, strict=False)
logger.info("Load pretrained resnet-18 parameters from: %s" %
model_file)
logger.info("Check md5sum of pretrained resnet-18 parameters: %s" %
md5sum(model_file))
def update_params(self):
model_file = self._hyper_params.get("pretrain_model_path", "")
if model_file != "":
state_dict = torch.load(model_file,
map_location=torch.device("cpu"))
if "model_state_dict" in state_dict:
state_dict = state_dict["model_state_dict"]
self.load_model_param(state_dict)
logger.info(
"Load pretrained {} parameters from: {} whose md5sum is {}".
format(self.__class__.__name__, model_file, md5sum(model_file)))
def update_params(self):
model_file = self._hyper_params["pretrain_model_path"]
if model_file != "":
try:
state_dict = torch.load(model_file,
map_location=torch.device("gpu"))
except:
state_dict = torch.load(model_file,
map_location=torch.device("cpu"))
self.load_state_dict(state_dict, strict=False)
logger.info("Load pretrained GoogLeNet parameters from: %s" %
model_file)
logger.info("Check md5sum of pretrained GoogLeNet parameters: %s" %
md5sum(model_file))
self.crop_pad = self._hyper_params['crop_pad']
self.pruned = self._hyper_params['pruned']
def update_params(self):
arch = "shufflenetv2_x1.0"
kwargs = self._hyper_params
# build module
self._model = _shufflenetv2(arch,
False,
True, [4, 8, 4], [24, 116, 232, 464, 1024],
fused_channls=[116, 232, 464],
**kwargs)
model_file = self._hyper_params["pretrain_model_path"]
if model_file != "":
state_dict = torch.load(model_file,
map_location=torch.device("cpu"))
self._model.load_state_dict(state_dict, strict=False)
logger.info("Load pretrained ShuffleNet parameters from: %s" %
model_file)
logger.info(
"Check md5sum of pretrained ShuffleNet parameters: %s" %
md5sum(model_file))
def update_params(self):
r"""
定义了 model 用配置文件提供的超参数更新超参数的行为
Returns:
"""
model_file = self._hyper_params.get("pretrain_model_path",
"") # 获取预训练模型文件位置
if model_file != "":
state_dict = torch.load(
model_file,
map_location=torch.device("cpu")) # 加载 model tensors
if "model_state_dict" in state_dict:
state_dict = state_dict["model_state_dict"]
self.load_model_param(state_dict)
logger.info(
"Load pretrained {} parameters from: {} whose md5sum is {}".
format(self.__class__.__name__, model_file,
md5sum(model_file)))
