-
Notifications
You must be signed in to change notification settings - Fork 2
/
gen_sequential_synth_data.py
217 lines (171 loc) · 8.71 KB
/
gen_sequential_synth_data.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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
import torch
import cv2
import imageio
import pickle
import os
from tqdm import tqdm
from numpy.linalg import norm
from Encoder_BH2MANO import Encoder_BH2MANO
from gen_3d_pose_flow import *
from manopth.manolayer import ManoLayer
from opendr.renderer import ColoredRenderer
from opendr.camera import ProjectPoints
from PIL import Image
def create_synth(verts, joints, skin_color, f, ss, tu, tv, rot, w, h, bg):
rn = ColoredRenderer()
R = cv2.Rodrigues(rot)[0]
verts = np.transpose(np.matmul(R, np.transpose(verts)))
joints = np.transpose(np.matmul(R, np.transpose(joints)))
verts_3d = verts
joints_3d = joints
verts = np.array([[ss, ss, 1], ] * 778) * verts
joints = np.array([[ss, ss, 1], ] * 21) * joints
verts = verts + np.array([[tu, tv, 0], ] * 778)
joints = joints + np.array([[tu, tv, 0], ] * 21)
umax = np.max(verts[:, 0])
umin = np.min(verts[:, 0])
vmax = np.max(verts[:, 1])
vmin = np.min(verts[:, 1])
if ((umin < 0.) or (vmin < 0.) or (umax > w) or (vmax > h)):
print('mesh outside')
verts[:, 2] = 10. + (verts[:, 2] - np.mean(verts[:, 2]))
verts[:, :2] = verts[:, :2] * np.expand_dims(verts[:, 2], 1)
rn.camera = ProjectPoints(v=verts, rt=np.zeros(3), t=np.array([0, 0, 0]), f=np.array([1, 1]),
c=np.array([0, 0]), k=np.zeros(5))
rn.frustum = {'near': 1., 'far': 20., 'width': w, 'height': h}
rn.set(v=verts, f=f, bgcolor=np.zeros(3))
rn.vc = np.ones((778, 3))
mask = rn.r.copy()
mask = mask[:, :, 0].astype(np.uint8)
rn.vc = skin_color
hand = rn.r.copy() * 255.
image = (1 - np.expand_dims(mask, 2)) * bg + np.expand_dims(mask, 2) * hand
image = image.astype(np.uint8)
image = Image.fromarray(image).resize((224, 224), Image.LANCZOS)
return image, mask, verts_3d, joints_3d, verts, joints
def get_ss_tu_tv(verts, joints, w, h) :
umax, umin, vmax, vmin = np.max(verts[:, 0]), np.min(verts[:, 0]), np.max(verts[:, 1]), np.min(verts[:, 1])
ss = w / ((random.uniform(1.7, 2.1) * np.max([umax - umin, vmax - vmin])))
verts, joints = np.array([[ss, ss, 1], ] * 778) * verts, np.array([[ss, ss, 1], ] * 21) * joints
umax, umin, vmax, vmin = np.max(verts[:, 0]), np.min(verts[:, 0]), np.max(verts[:, 1]), np.min(verts[:, 1])
hand_width = umax + umin
hand_length = vmax + vmin
tu = (w - hand_width) / 2. + umin + hand_width / 2. + random.uniform(15,15)
tv = (h - hand_length) / 2. + vmin + hand_length / 2. + random.uniform(-50,45)
rot_ind = random.randint(0,1)
if rot_ind > 0.5 :
angle = random.uniform(-np.pi, np.pi)
axis = np.array([random.uniform(-1., 1.) for _ in range(3)])
axis[random.randint(0, 2)] = 1.
axis /= np.linalg.norm(axis)
rot = angle * axis
else :
angle = np.array([random.uniform(0.1, 0.3) for _ in range(3)])
rot = np.array([-np.pi/3-angle[0], -np.pi/3-angle[1], np.pi/3+angle[2]])
return ss, tu, tv, rot
def get_bg(bg_shape, size):
bg_cent_x = random.randint(size / 2, bg_shape[0] - size / 2)
bg_cent_y = random.randint(size / 2, bg_shape[1] - size / 2)
return bg_cent_x, bg_cent_y
def get_hand_colors() :
colors = []
for i in range(0, 27):
f = open('meshes_colored/%d.obj' % i)
cont = f.readlines()
f.close()
col = []
for x in cont:
a = x[:len(x) - 1].split(" ")
if (a[0] == 'v'):
col.append(np.array([float(a[4]), float(a[5]), float(a[6])]))
col = np.expand_dims(np.vstack(col), 0)
colors.append(col)
return np.vstack(colors)
def main() :
mano_layer = ManoLayer(mano_root='mano/models', use_pca=True, ncomps=6, flat_hand_mean=False, side='right')
gpu = True
num_flows = 2
flow_length = 10
pd = np.load('normalized_bh_inMANOorder.npy') ## Need to Download it
model_path = 'BH2MANO_model/model_BH2MANO.pth'
bg_path = 'BG_data/'
bg_files = os.listdir(os.path.join(os.getcwd(), bg_path))
data_dir = 'results/'
enc_layer = Encoder_BH2MANO()
enc_layer.load_state_dict(torch.load(model_path))
if gpu:
enc_layer.cuda()
mano_layer.cuda()
mano_layer.eval()
enc_layer.eval()
colors = get_hand_colors()
gt = {}
gt['joints_2d'] = []
gt['cam_params'] = []
gt['joints_3d'] = []
gt['verts_3d'] = []
for frame_idx in tqdm(range(num_flows)):
pose_flow = gen_3d_pose_flow(pd.reshape((pd.shape[0], pd.shape[1] * pd.shape[2])), flow_length=flow_length)
samples = pose_flow
input_pose_joints_ = torch.tensor(samples).float().cuda()
input_pose_joints = input_pose_joints_.view(input_pose_joints_.size(0), -1)
pose_param,shape_param = enc_layer(input_pose_joints)
pose_param[:, :3] *= 0.
pose_param[:, 0] += np.pi
shape_param = torch.rand(1, 10).expand(pose_param.size(0), -1).cuda().float() * 4. - 2.
hand_verts, hand_joints = mano_layer(pose_param.clone(), shape_param.clone())
hand_verts = hand_verts.cpu().detach().numpy() / 1000.
hand_joints = hand_joints.cpu().detach().numpy() / 1000.
f = mano_layer.th_faces.cpu().detach().numpy()
color = colors[random.randint(0, 26)]
size = 224
w, h = size, size
# flow_length = pose_flow.shape[0]
assert flow_length == pose_flow.shape[0]
ss, tu, tv, rot = get_ss_tu_tv(hand_verts[0], hand_joints[0], w, h)
ss_end, tu_end, tv_end, rot_end = get_ss_tu_tv(hand_verts[-1], hand_joints[-1], w, h)
rot_var_speed = random.uniform(0, 0.6) # random rotation speed
## Get Background
while True :
bg_orig = imageio.imread(os.path.join(bg_path, random.choice(bg_files)))
if (bg_orig.shape[0] > size) and (bg_orig.shape[1] > size) : # get background that is large enough
break
bg_cent_x, bg_cent_y = get_bg(bg_orig.shape, size)
bg_cent_end_x, bg_cent_end_y = get_bg(bg_orig.shape, size)
bg = bg_orig[bg_cent_x - int(size / 2): bg_cent_x + int(size / 2),
bg_cent_y - int(size / 2): bg_cent_y + int(size / 2), :]
## Collect GTs
images = []
masks = []
joints_2d = np.zeros((flow_length, 42))
cam_params = np.zeros((flow_length, 27))
joints_3d = np.zeros((flow_length, 21, 3))
verts_3d = np.zeros((flow_length, 778, 3))
for i in range(flow_length):
img, mask, vert_3d, joint_3d, vert, joint = create_synth(hand_verts[i], hand_joints[i], color, f, ss, tu, tv, rot, w, h, bg)
images.append(img)
masks.append(mask * 255)
cam_params[i, :] = np.concatenate([np.array([1., ss, tu, tv]), rot,
pose_param[i, 3:].detach().cpu().numpy(),
shape_param[i].detach().cpu().numpy()], 0)
joints_2d[i, :] = joint[:,:2].reshape((42))
joints_3d[i, :, :] = joint_3d
verts_3d[i, :, :] = vert_3d
ss = ss + (ss_end - ss) / flow_length * 0.5
tu = tu + (tu_end - tu) / flow_length * 0.2
tv = tv + (tv_end - tv) / flow_length * 0.2
rot = rot + (rot_end - rot) / flow_length * rot_var_speed
bg_cent_x = int(bg_cent_x + (bg_cent_end_x - bg_cent_x) / flow_length)
bg_cent_y = int(bg_cent_y + (bg_cent_end_y - bg_cent_y) / flow_length)
bg = bg_orig[bg_cent_x - int(size / 2):bg_cent_x + int(size / 2),
bg_cent_y - int(size / 2):bg_cent_y + int(size / 2), :]
gt['joints_2d'].append(joints_2d)
gt['joints_3d'].append(joints_3d)
gt['verts_3d'].append(verts_3d)
gt['cam_params'].append(cam_params)
imageio.mimsave(data_dir + 'gifs/%s.gif' % (frame_idx), images)
imageio.mimsave(data_dir + 'masks/synth_%s_mask.gif' % (frame_idx), masks)
with open(data_dir + 'ground_truths.pickle', 'wb') as fo:
pickle.dump(gt, fo, protocol=pickle.HIGHEST_PROTOCOL)
if __name__=="__main__":
main()