os.path.join(data_dir, "cifar10_val.rec"))
download_file('http://data.mxnet.io/data/cifar10/cifar10_val.rec', fnames[1])
download_file('http://data.mxnet.io/data/cifar10/cifar10_train.rec', fnames[0])
return fnames
if __name__ == '__main__':
# download data
(train_fname, val_fname) = download_cifar10()
# parse args
parser = argparse.ArgumentParser(description="train cifar10",
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
fit.add_fit_args(parser)
data.add_data_args(parser)
data.add_data_aug_args(parser)
data.set_data_aug_level(parser, 2)
parser.set_defaults(
# network
network = 'resnet',
num_layers = 110,
# data
data_train = train_fname,
data_val = val_fname,
num_classes = 10,
num_examples = 50000,
image_shape = '3,28,28',
pad_size = 4,
# train
batch_size = 512,
num_epochs = 300,
if __name__ == "__main__":
# parse args
parser = argparse.ArgumentParser(description="fine-tune a dataset",
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
train = fit.add_fit_args(parser)
data.add_data_args(parser)
aug = data.add_data_aug_args(parser)
parser.add_argument('--pretrained-model', type=str,
help='the pre-trained model')
parser.add_argument('--layer-before-fullc', type=str, default='flatten0',
help='the name of the layer before the last fullc layer')
# use less augmentations for fine-tune
data.set_data_aug_level(parser, 1)
set_imagenet_aug(parser)
# use a small learning rate and less regularizations
parser.set_defaults(image_shape='3,224,224', num_epochs=30,
lr=.01, lr_step_epochs='20', wd=0, mom=0)
args = parser.parse_args()
# load pretrained model
dir_path = os.path.dirname(os.path.realpath(__file__))
(prefix, epoch) = modelzoo.download_model(
args.pretrained_model, os.path.join(dir_path, 'model'))
if args.load_epoch is not None:
(prefix, epoch) = (args.model_prefix, args.load_epoch)
logging.info(prefix)
logging.info(epoch)
def start_new_training():
# binding signals
SUPPORTED_SIGNALS = (
signal.SIGINT,
signal.SIGTERM,
)
for signum in SUPPORTED_SIGNALS:
try:
signal.signal(signum, signal_handler)
logger.info("Bind signal '%s' success to %s", signum,
signal_handler)
except Exception as identifier:
logger.warning("Bind signal '%s' failed, err: %s", signum,
identifier)
try:
# parse args
parser = argparse.ArgumentParser(
description="train imagenet-1k",
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
fit.add_fit_args(parser)
data.add_data_args(parser)
data.add_data_aug_args(parser)
# use a large aug level
data.set_data_aug_level(parser, 3)
parser.set_defaults(
# network
network='resnet',
num_layers=50,
# data
num_classes=10,
num_examples=60000,
image_shape='3,28,28',
min_random_scale=1, # if input image has min size k, suggest to use
# 256.0/x, e.g. 0.533 for 480
# train
num_epochs=80,
lr_step_epochs='30,60',
dtype='float32',
batch_size=32)
args = parser.parse_args()
# AVA-SDK new an Instance
train_ins = train.TrainInstance()
# add CALLBACK
batch_end_cb = train_ins.get_monitor_callback(
"mxnet", batch_size=args.batch_size, batch_freq=10)
args.batch_end_callback = batch_end_cb
# load network
from importlib import import_module
net = import_module('symbols.' + args.network)
sym = net.get_symbol(**vars(args))
# train
fit.fit(args, sym, data.get_rec_iter)
logger.info("training finish")
err_msg = ""
if train_ins == None:
return
train_ins.done(err_msg=err_msg)
except Exception as err:
err_msg = "training failed, err: %s" % (err)
logger.info(err_msg)
traceback.print_exc(file=sys.stderr)
if train_ins == None:
return
train_ins.done(err_msg=err_msg)