|
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__": |
|
|
|
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() |
|
|
|
|
|
with open(args.config, "r") as f: |
|
config = yaml.load(f, Loader=yaml.FullLoader) |
|
|
|
|
|
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() |
|
|