parser.add_argument('--render', action='store_true')
parser.add_argument('--mpc_horizon', type=int, default=15)
parser.add_argument('--num_random_action_selection', type=int, default=4096)
parser.add_argument('--nn_layers', type=int, default=1)
parser.add_argument('--CEM_mode', action="store_true")
args = parser.parse_args()
data_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'data')
exp_name = '{0}_{1}_{2}'.format(
args.env, args.question,
args.exp_name if args.exp_name else time.strftime("%d-%m-%Y_%H-%M-%S"))
exp_dir = os.path.join(data_dir, exp_name)
assert not os.path.exists(exp_dir),\
'Experiment directory {0} already exists. Either delete the directory, or run the experiment with a different name'.format(exp_dir)
os.makedirs(exp_dir, exist_ok=True)
logger.setup(exp_name, os.path.join(exp_dir, 'log.txt'), 'debug')
env = {'HalfCheetah': HalfCheetahEnv()}[args.env]
mbrl = ModelBasedRL(
env=env,
render=args.render,
mpc_horizon=args.mpc_horizon,
num_random_action_selection=args.num_random_action_selection,
nn_layers=args.nn_layers,
CEM_mode=args.CEM_mode)
run_func = {
'q1': mbrl.run_q1,
'q2': mbrl.run_q2,
'q3': mbrl.run_q3
# add common library
from logger import logger, log
logger.setup('./logs', name='efficientDet-d5-cutmix-sgd')
from lib import *
from config import config
from dataset import WheatDataset, get_train_transforms, get_valid_transforms
from utils import seed_everything, read_csv, kfold
from trainer import Trainner, collate_fn
from efficientdet_master.effdet import get_efficientdet_config, EfficientDet, DetBenchTrain
from efficientdet_master.effdet.efficientdet import HeadNet
def get_net():
# config = get_efficientdet_config('tf_efficientdet_d7')
# net = EfficientDet(config, pretrained_backbone=False)
# checkpoint = torch.load('./input/efficientdet/tf_efficientdet_d7-f05bf714.pth') #D7
net_config = get_efficientdet_config('tf_efficientdet_d5')
net = EfficientDet(net_config, pretrained_backbone=config.use_pretrained)
checkpoint = torch.load(
'./input/efficientdet/tf_efficientdet_d5-ef44aea8.pth') #D5
net.load_state_dict(checkpoint)
net_config.num_classes = 1
net_config.image_size = 1024
net.class_net = HeadNet(net_config,
num_outputs=net_config.num_classes,
norm_kwargs=dict(eps=.001, momentum=.01))
return DetBenchTrain(net, net_config)
