-
Notifications
You must be signed in to change notification settings - Fork 0
/
training.py
36 lines (29 loc) · 1.12 KB
/
training.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
import os
import sys
import argparse
from random import seed
import numpy as np
import torch
from src.config import load_config, load_scene_list, load_invalid_frames
from src.vfuse import VFUSE
# argument parsing
parser = argparse.ArgumentParser(description="V-FUSE Network training.")
parser.add_argument("--config_path", type=str, help="Configuration path.")
parser.add_argument("--dataset", type=str, help='Current dataset being used.', choices=["scannet", "replica", "dtu", "tnt"], required=True)
ARGS = parser.parse_args()
def main():
#### Load Configuration ####
cfg = load_config(os.path.join(ARGS.config_path, f"{ARGS.dataset}.yaml"))
cfg["mode"] = "training"
# set random seed
torch.manual_seed(cfg["seed"])
seed(cfg["seed"])
np.random.seed(cfg["seed"])
#### Load Scene Lists ####
ts = load_scene_list(os.path.join(cfg["scene_list_path"], "training.txt"))
vs = load_scene_list(os.path.join(cfg["scene_list_path"], "validation.txt"))
#### TRAINING ####
pipeline = VFUSE(cfg, training_scenes=ts, validation_scenes=vs)
pipeline.training()
if __name__ == '__main__':
main()