Spaces:
Running
Running
import cv2 | |
import yaml | |
import argparse | |
import os | |
from torch.utils.data import DataLoader | |
from datasets.gl3d_dataset import GL3DDataset | |
from trainer import Trainer | |
from trainer_single_norel import SingleTrainerNoRel | |
from trainer_single import SingleTrainer | |
if __name__ == '__main__': | |
# add argument parser | |
parser = argparse.ArgumentParser() | |
parser.add_argument('--config', type=str, default='./configs/config.yaml') | |
parser.add_argument('--dataset_dir', type=str, default='/mnt/nvme2n1/hyz/data/GL3D') | |
parser.add_argument('--data_split', type=str, default='comb') | |
parser.add_argument('--is_training', type=bool, default=True) | |
parser.add_argument('--job_name', type=str, default='') | |
parser.add_argument('--gpu', type=str, default='0') | |
parser.add_argument('--start_cnt', type=int, default=0) | |
parser.add_argument('--stage', type=int, default=1) | |
args = parser.parse_args() | |
# load global config | |
with open(args.config, 'r') as f: | |
config = yaml.load(f, Loader=yaml.FullLoader) | |
# setup dataloader | |
dataset = GL3DDataset(args.dataset_dir, config['network'], args.data_split, is_training=args.is_training) | |
data_loader = DataLoader(dataset, batch_size=2, shuffle=True, num_workers=4) | |
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu | |
if args.stage == 1: | |
trainer = SingleTrainerNoRel(config, f'cuda:0', data_loader, args.job_name, args.start_cnt) | |
elif args.stage == 2: | |
trainer = SingleTrainer(config, f'cuda:0', data_loader, args.job_name, args.start_cnt) | |
elif args.stage == 3: | |
trainer = Trainer(config, f'cuda:0', data_loader, args.job_name, args.start_cnt) | |
else: | |
raise NotImplementedError() | |
trainer.train() | |