/
rgb_multiview_hand.py
168 lines (149 loc) · 5.08 KB
/
rgb_multiview_hand.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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
"""Example with fitting a colored hand mesh model to multiple images."""
import datetime
import glob
import json
import os
import time
import cv2
from imageio import imread, imwrite
import matplotlib.pyplot as plt
import numpy as np
from typing_extensions import Literal
import deodr
from deodr import read_obj
from deodr.mesh_fitter import MeshRGBFitterWithPoseMultiFrame
dl_library_type = Literal["pytorch", "tensorflow", "none"]
def run(
dl_library: dl_library_type = "pytorch",
plot_curves: bool = False,
save_images: bool = False,
display: bool = True,
max_iter: int = 400,
) -> None:
file_folder = os.path.dirname(__file__)
hand_images = [
imread(file).astype(np.double) / 255
for file in glob.glob(os.path.join(deodr.data_path, "./hand_multiview/*.jpg"))
]
nb_frames = len(hand_images)
obj_file = os.path.join(deodr.data_path, "hand.obj")
faces, vertices = read_obj(obj_file)
default_color = np.array([0.4, 0.3, 0.25]) * 1.5
default_light_directional = -np.array([0.1, 0.5, 0.4])
default_light_ambient = 0.6
# default_light = {'directional':np.array([0.0,0.0,0.0]),'ambient':np.array([0.6])}
euler_init = np.row_stack(
[np.array([0, yrot, 0]) for yrot in np.linspace(-0.5, 0.5, 3)]
)
vertices = vertices - np.mean(vertices, axis=0)
t_init = np.array([0, -0.2, 0.2])
translation_init = np.tile(t_init[None, :], [nb_frames, 1])
# centering vertices
hand_fitter = MeshRGBFitterWithPoseMultiFrame(
vertices,
faces,
default_color=default_color,
default_light_directional=default_light_directional,
default_light_ambient=default_light_ambient,
update_lights=True,
update_color=True,
euler_init=euler_init,
translation_init=translation_init,
cregu=2000,
)
# handFitter = MeshRGBFitter(vertices,faces,default_color,default_light,
# update_lights = True, update_color= True,cregu=1000)
hand_fitter.reset()
max_iter = 300
hand_image = hand_images[0]
background_color = np.median(
np.row_stack(
(
hand_image[:10, :10, :].reshape(-1, 3),
hand_image[-10:, :10, :].reshape(-1, 3),
hand_image[-10:, -10:, :].reshape(-1, 3),
hand_image[:10, -10:, :].reshape(-1, 3),
)
),
axis=0,
)
background_color = np.array([0, 0, 0])
hand_fitter.set_images(hand_images)
hand_fitter.set_background_color(background_color)
energies = []
durations = []
start = time.time()
iter_folder = os.path.join(file_folder, "./iterations/depth")
if not os.path.exists(iter_folder):
os.makedirs(iter_folder)
for niter in range(max_iter):
energy, image, diff_image = hand_fitter.step(check_gradient=False)
energies.append(energy)
durations.append(time.time() - start)
if display or save_images:
combined_image = np.column_stack(
(
np.row_stack(hand_images),
image,
np.tile(np.minimum(diff_image, 1)[:, :, None], (1, 1, 3)),
)
)
if display:
cv2.imshow(
"animation",
cv2.resize(combined_image[:, :, ::-1], None, fx=1, fy=1),
)
if save_images:
imwrite(
os.path.join(iter_folder, f"hand_iter_{niter}.png"),
(combined_image * 255).astype(np.uint8),
)
cv2.waitKey(1)
# save convergence curve
with open(
os.path.join(
iter_folder,
f'rgb_image_fitting_result_{str(datetime.datetime.now()).replace(":", "_")}.json',
),
"w",
) as f:
json.dump(
{
"label": f"{dl_library} {datetime.datetime.now()}",
"durations": durations,
"energies": energies,
},
f,
indent=4,
)
# compare with previous runs
if plot_curves:
plt.figure()
for file in glob.glob(
os.path.join(iter_folder, "rgb_image_fitting_result_*.json")
):
with open(file, "r") as fp:
json_data = json.load(fp)
plt.plot(
json_data["durations"],
json_data["energies"],
label=json_data["label"],
)
plt.xlabel("duration in seconds")
plt.ylabel("energies")
plt.legend()
plt.figure()
for file in glob.glob(
os.path.join(iter_folder, "rgb_image_fitting_result_*.json")
):
with open(file, "r") as fp:
json_data = json.load(fp)
plt.plot(json_data["energies"], label=json_data["label"])
plt.xlabel("iterations")
plt.ylabel("energies")
plt.legend()
plt.show()
if __name__ == "__main__":
display = True
save_images = False
run(dl_library="none", plot_curves=True, display=display, save_images=save_images)