-
Notifications
You must be signed in to change notification settings - Fork 1
/
inference.py
143 lines (118 loc) · 4.25 KB
/
inference.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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
import cv2
cv2.setNumThreads(0)
import argparse
import os
import sys
from pathlib import Path
import gdown
import torch
import tqdm
import yaml
from einops import rearrange
project_dir = Path(__file__).parent.resolve()
sys.path.insert(0, str(project_dir / "src"))
from inv3d_model.models import model_factory
from inv3d_util.image import scale_image
from inv3d_util.load import load_image, save_image, save_npz
from inv3d_util.mapping import apply_map_torch
from inv3d_util.misc import to_numpy_image, to_numpy_map
from inv3d_util.path import list_dirs
model_sources = yaml.safe_load((project_dir / "models.yaml").read_text())
def inference(model_name: str, dataset: str, gpu: int, output_shape: tuple[int, int]):
os.environ["CUDA_VISIBLE_DEVICES"] = str(gpu)
model_url = model_sources[model_name]
model_path = Path(
gdown.cached_download(
url=model_url, path=project_dir / f"models/{model_name}.ckpt"
)
)
model = model_factory.load_from_checkpoint(model_name.split("@")[0], model_path)
model.to("cuda")
model.eval()
input_dir = project_dir / "input" / dataset
output_dir = project_dir / "output" / f"{dataset} - {model_name}"
output_dir.mkdir(exist_ok=True)
image_paths = list(input_dir.glob("image_*.*"))
for image_path in tqdm.tqdm(image_paths, "Unwarping images"):
sample_name = image_path.stem.removeprefix("image_")
# prepare image
image_original = load_image(image_path)
image = scale_image(
image_original, resolution=model.dataset_options["resolution"]
)
image = rearrange(image, "h w c -> () c h w")
image = image.astype("float32") / 255
image = torch.from_numpy(image)
image = image.to("cuda")
model_kwargs = {"image": image}
# prepare template
if "template" in model_name:
[template_path] = list(input_dir.glob(f"template_{sample_name}.*"))
template_original = load_image(template_path)
template = scale_image(
template_original, resolution=model.dataset_options["resolution"]
)
template = rearrange(template, "h w c -> () c h w")
template = template.astype("float32") / 255
template = torch.from_numpy(template)
template = template.to("cuda")
model_kwargs["template"] = template
# inference model
out_bm = model(**model_kwargs).detach().cpu()
# unwarp input
image_original = rearrange(image_original, "h w c -> () c h w")
image_original = image_original.astype("float32") / 255
image_original = torch.from_numpy(image_original)
norm_image = apply_map_torch(
image=image_original, bm=out_bm, resolution=output_shape
)
# export results
save_image(
output_dir / f"unwarped_{sample_name}.png",
to_numpy_image(norm_image),
override=True,
)
save_npz(
output_dir / f"bm_{sample_name}.npz", to_numpy_map(out_bm), override=True
)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--model",
type=str,
choices=list(model_sources.keys()),
required=True,
help="Select the model and the dataset used for training.",
)
parser.add_argument(
"--dataset",
type=str,
choices=list(map(lambda x: x.name, list_dirs(project_dir / "input"))),
required=True,
help="Selects the inference dataset. All folders in the input directory can be selected.",
)
parser.add_argument(
"--gpu",
type=int,
required=True,
help="The index of the GPU to use for inference.",
)
parser.add_argument(
"--output_width",
type=int,
default=1700,
help="Defines the width of the output document in pixels.",
)
parser.add_argument(
"--output_height",
type=int,
default=2200,
help="Defines the height of the output document in pixels.",
)
args = parser.parse_args()
inference(
model_name=args.model,
dataset=args.dataset,
gpu=args.gpu,
output_shape=(args.output_height, args.output_width),
)