def __init__(self, args):
self.args = args
if self.args.test_save_dir is None:
self.save_dir = tools.set_save_dir(args)
else:
self.save_dir = self.args.test_save_dir
self.test_loader = inputs.get_dataloader(self.args)
model = getattr(models, self.args.model)(self.args)
self.model = model.cuda()
self.load_best_model()
def __init__(self, args, is_trainval=True, writer=None, val_loader=None):
super(Validator, self).__init__(args)
self.is_trainval = is_trainval
self.save_dir = tools.set_save_dir(args)
self.writer = writer
if is_trainval:
self.val_loader = val_loader
else:
self.val_loader = inputs.get_dataloader(self.args)
assert (writer is None) and (val_loader is None)
print(self.args)
self.visualizer = visualizer.Visualizer(args, "val", self.save_dir, self.writer)
def __init__(self, args):
super(Trainer, self).__init__(args)
#### 0. Setup
self.save_dir = tools.set_save_dir(args)
with open(os.path.join(self.save_dir, "args.json"), "w") as j:
json.dump(vars(args), j)
#### 1. Models
model = getattr(models, args.model)(args)
print(
"Model param nums: ",
sum(p.numel() for p in model.parameters() if p.requires_grad),
)
self.model = model.cuda()
#### 2. Opt
self.optimizer = opts.get_optimizer(args, self.model)
self.scheduler = None
if self.args.lr_scheduler is not None:
self.scheduler = opts.get_scheduler(args, self.optimizer)
#### 3. Data
if args.augment is not None:
augmentation = getattr(augmentations, args.augment)
else:
augmentation = None
self.train_loader, self.val_loader = inputs.get_dataloader(
args, transform=augmentation)
#### 4. Logger
self.writer = writer.Writer(log_dir=self.save_dir)
self.logger = logger.Logger()
self.logger.open(os.path.join(self.save_dir, "log.train.txt"),
mode="a")
self.logger.write("\n>> Pytorch version: {}".format(torch.__version__))
self.logger.write("\n>> Args: {}".format(args))
# Validator
self.validator = Validator(
args,
is_trainval=True,
writer=self.writer,
val_loader=self.val_loader,
)
def __init__(self, args, is_trainval=True, writer=None, val_loader=None):
self.args = args
self.is_trainval = is_trainval
self.save_dir = tools.set_save_dir(args)
self.writer = writer
# self.prob_ths = prob_ths
if is_trainval:
self.val_loader = val_loader
else:
print(args)
assert (writer is None) and (val_loader is None)
self.val_loader = my_input.get_dataloader(self.args)
self.visualizer = visualizer.Visualizer(args, "val", self.save_dir,
self.writer)
def __init__(self, args):
self.args = args
#### 0. Setup
self.save_dir = tools.set_save_dir(args)
with open(os.path.join(self.save_dir, "args.json"), "w") as j:
json.dump(vars(args), j)
#### 1. Data
# TODO: augmentation
augmentation = getattr(augmentations, args.augment)
self.train_loader, self.val_loader = my_input.get_dataloader(
args, transform=augmentation)
#### 2. Model
model = models.PENetClassifier(**vars(args))
model.load_pretrained(PRETRAINED_WEIGHTS, "0")
self.model = model.cuda()
#### 3. Opt
self.optimizer = opts.get_optimizer(args, self.model)
self.scheduler = None
if self.args.lr_scheduler is not None:
self.scheduler = opts.get_scheduler(args, self.optimizer)
#### 4. Logger
self.writer = writer.Writer(log_dir=self.save_dir)
self.logger = logger.Logger()
self.logger.open(os.path.join(self.save_dir, "log.train.txt"),
mode="a")
self.logger.write("\n>> Pytorch version: {}".format(torch.__version__))
self.logger.write("\n>> Args: {}".format(args))
# self.visualizer = visualizer.Visualizer(
# args, "train", self.save_dir, self.writer
# )
# Validator
self.validator = Validator(
args,
is_trainval=True,
writer=self.writer,
val_loader=self.val_loader,
)