-
Notifications
You must be signed in to change notification settings - Fork 4
/
make_movie.py
46 lines (41 loc) · 1.49 KB
/
make_movie.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
37
38
39
40
41
42
43
44
45
46
from lib.utils import LatentSpaceExplorer
import torch
import numpy as np
import argparse
def parse_args():
parser = argparse.ArgumentParser(
description="Train a Neural ODE model for semantic image manipulation"
)
parser.add_argument(
"--dataset", default="ffhq", choices=["ffhq", "cub", "scenes"],
)
parser.add_argument(
"--w-path", default=None, type=str, help="Path to precomputed style vectors."
)
parser.add_argument("--nrow", default=2, type=int)
parser.add_argument("--depth", type=int, default=1)
parser.add_argument("--attribute", type=str)
parser.add_argument("--format", type=str, default="mp4", choices=["gif", "mp4"])
parser.add_argument("--duration", type=float, default=2.0)
parser.add_argument("--hi", type=float, default=4.0)
parser.add_argument("--steps", type=int, default=20)
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_args()
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
explorer = LatentSpaceExplorer(args.dataset, device=device)
if args.w_path is not None:
w = torch.load(args.w_path).to(device)
else:
w = explorer.generate_latent(args.nrow ** 2)
explorer.get_flow_frames(
w,
hi=args.hi,
steps=args.steps,
attr_name=args.attribute,
save_gif=True,
save_as_mp4=(args.format == "mp4"),
duration=args.duration,
depth=args.depth,
)