Spaces:
Running on Zero
Running on Zero
File size: 10,807 Bytes
2680bd5 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 | import numpy as np
import json
from ..constant import JOINT_NAME_INDEX_MAP
from .single_motioncode import MotionCoder, split_events, split_contact_events
def compute_topk_closest_vectors(A, B, k=10):
'''
A: (T, 3)
B: (T, n, 3)
return: (T, k, 3)
'''
A_expanded = A[:, np.newaxis, :] # (T, 1, 3)
diff = B - A_expanded # (T, n, 3)
dists = np.linalg.norm(diff, axis=2) # (T, n)
topk_indices = np.argpartition(dists, kth=k, axis=1)[:, :k] # (T, k)
row_indices = np.arange(A.shape[0])[:, np.newaxis] # (T, 1)
sorted_order = np.argsort(dists[row_indices, topk_indices], axis=1)
topk_sorted_indices = topk_indices[row_indices, sorted_order]
C = diff[row_indices, topk_sorted_indices] # (T, k, 3)
return C
def compute_closest_distances(A, B):
'''
A: (T, n, 3)
B: (T, n, 3)
return
- C: (T, n)
- C_vec: (T, n, 3)
'''
T, n, _ = A.shape
A_exp = A[:, :, np.newaxis, :] # (T, n, 1, 3)
B_exp = B[:, np.newaxis, :, :] # (T, 1, n, 3)
diff = A_exp - B_exp # (T, n, n, 3)
dists = np.linalg.norm(diff, axis=-1) # (T, n, n)
C = np.min(dists, axis=2) # (T, n)
indices = np.argmin(dists, axis=2) # (T, n)
C_vec = np.empty((T, n, 3))
for t in range(T):
for i in range(n):
j = indices[t, i]
C_vec[t, i] = -A[t, i] + B[t, j]
return C, C_vec
def topk_smallest_elements(C, k):
topk_part = np.partition(C, kth=k-1, axis=1)[:, :k]
D = np.sort(topk_part, axis=1)
return D
class BihandMotionCoder(object):
def __init__(self, motion:np.ndarray):
self.lft_motion = motion[:, 0] # (T, J, 3)
self.rgt_motion = motion[:, 1] # (T, J, 3)
# print(f"lft motion z mean: {np.mean(self.lft_motion[:, :, 2])}")
# print(f"rgt motion z mean: {np.mean(self.rgt_motion[:, :, 2])}")
self.lft_motioncode = MotionCoder(self.lft_motion, isright=False)
self.rgt_motioncode = MotionCoder(self.rgt_motion, isright=True)
self.tip_index = [
JOINT_NAME_INDEX_MAP['thumb_tip'],
JOINT_NAME_INDEX_MAP['index_tip'],
JOINT_NAME_INDEX_MAP['middle_tip'],
JOINT_NAME_INDEX_MAP['ring_tip'],
JOINT_NAME_INDEX_MAP['pinky_tip'],
]
self.unit_len = self.lft_motioncode.unit_len
def get_finger_finger_distance(self):
finger_finger_distance = []
tips = ['thumb_tip', 'index_tip', 'middle_tip', 'ring_tip', 'pinky_tip']
for i in range(len(tips)):
for j in range(len(tips)):
joint_pair = [JOINT_NAME_INDEX_MAP[tips[i]], JOINT_NAME_INDEX_MAP[tips[j]]]
dist = np.linalg.norm(self.lft_motion[:, joint_pair[0], :] - self.rgt_motion[:, joint_pair[1], :], axis=1) # (T,)
finger_finger_distance.append((
('left_' + tips[i], "right_" + tips[j]),
dist
))
# if tips[i] == 'index_tip' and tips[j] == 'index_tip':
# print(f"{joint_pair}:\n{dist}")
self.finger_finger_distance = finger_finger_distance
def get_finger_palm_distance(self):
finger_palm_distance, palm_finger_distance = [], []
tips = ['thumb_tip', 'index_tip', 'middle_tip', 'ring_tip', 'pinky_tip']
for i in range(len(tips)):
closest_vectors = compute_topk_closest_vectors(
self.lft_motion[:, JOINT_NAME_INDEX_MAP[tips[i]], :], # (T, 3)
self.rgt_motioncode.palm_points_glob, # (T, n, 3)
k=5
) # (T, k, 3)
closest_mean_dist = np.mean(np.linalg.norm(closest_vectors, axis=2), axis=1) # (T,)
finger_palm_distance.append((
('left_' + tips[i], "right_palm"),
closest_mean_dist
))
for i in range(len(tips)):
closest_vectors = compute_topk_closest_vectors(
self.rgt_motion[:, JOINT_NAME_INDEX_MAP[tips[i]], :], # (T, 3)
self.lft_motioncode.palm_points_glob, # (T, n, 3)
k=5
)
closest_mean_dist = np.mean(np.linalg.norm(closest_vectors, axis=2), axis=1)
palm_finger_distance.append((
('right_' + tips[i], "left_palm"),
closest_mean_dist
))
self.finger_palm_distance = finger_palm_distance
self.palm_finger_distance = palm_finger_distance
def get_palm_palm_distance(self):
closest_dist, closest_vec = compute_closest_distances(
self.lft_motioncode.palm_points_glob,
self.rgt_motioncode.palm_points_glob
)
self.palm_palm_vec = np.mean(self.rgt_motioncode.palm_points_glob, axis=1) - np.mean(self.lft_motioncode.palm_points_glob, axis=1) # (T, 3)
# print(f"palm_palm_vec: {self.palm_palm_vec}")
closest_dist_topk = topk_smallest_elements(closest_dist, k=30)
closest_dist_mean = np.mean(closest_dist_topk, axis=-1) # (T,)
self.palm_palm_distance = [(
('left_palm', 'right_palm'),
closest_dist_mean
)]
def extract_feats(self):
# print("GET LEFT MOTION FEATS")
self.lft_motioncode.extract_feats()
# print("GET RIGHT MOTION FEATS")
self.rgt_motioncode.extract_feats()
self.get_finger_finger_distance()
self.get_finger_palm_distance()
self.get_palm_palm_distance()
def split_finger_finger_events(self):
contact_thre = 0.020
self.finger_finger_distance_events = dict()
for pair, spacing in self.finger_finger_distance:
events = split_contact_events(spacing, contact_thre)
if len(events) > 0:
events = [
{k: v for k, v in x.items() if k != 'direction'}
for x in events
]
self.finger_finger_distance_events[', '.join(pair)] = events
def split_finger_palm_events(self):
spacing_thre = [
(0, 0.025, 'Contact'),
(0.025, 0.035, 'Near'),
(0.035, 100, 'Far')
]
vel_thre = [
(0, 0.025, 'Slow'),
(0.025, 0.07, 'Medium'),
(0.07, 100, 'Fast')
]
self.finger_palm_distance_events = dict()
for pair, spacing in self.finger_palm_distance:
events = split_events(
spacing, spacing_thre, vel_thre,
delta_thre=0.03,
min_duration=6,
diff_state=True,
unit_len=self.unit_len,
)
if len(events) > 0:
events = [
{k: v for k, v in x.items() if k != 'direction'}
for x in events
]
self.finger_palm_distance_events[', '.join(pair)] = events
for pair, spacing in self.palm_finger_distance:
events = split_events(spacing, spacing_thre, vel_thre)
if len(events) > 0:
events = [
{k: v for k, v in x.items() if k != 'direction'}
for x in events
]
self.finger_palm_distance_events[', '.join(pair)] = events
def split_palm_palm_events(self):
spacing_thre = [
(0, 0.04, 'Contact'),
(0.04, 0.08, 'Near'),
(0.08, 0.12, 'Medium'),
(0.12, 100, 'Far')
]
vel_thre = [
(0, 0.025, 'Slow'),
(0.025, 0.07, 'Medium'),
(0.07, 100, 'Fast')
]
self.palm_palm_distance_events = dict()
for pair, spacing in self.palm_palm_distance:
events = split_events(
x=spacing,
x_intervals=spacing_thre,
v_intervals=vel_thre,
delta_thre=0.05,
min_duration=6,
diff_state=True,
unit_len=self.unit_len,
)
if len(events) > 0:
events = [
{k: v for k, v in x.items() if k != 'direction'}
for x in events
]
self.palm_palm_distance_events[', '.join(pair)] = events
relation_thre = [
[
(-100, -0.02, 'right hand is to the LEFT of the left hand.'),
(-0.02, 0.02, 'right hand is ALIGNED with the left hand.'),
(0.02, 100, 'right hand is to the RIGHT of the left hand.')
],
[
(-100, -0.02, 'right hand is to the BACK of the left hand.'),
(-0.02, 0.02, 'right hand is ALIGNED with the left hand.'),
(0.02, 100, 'right hand is to the FRONT of the left hand.')],
[
(-100, -0.02, 'right hand is to the DOWN of the left hand.'),
(-0.02, 0.02, 'right hand is ALIGNED with the left hand.'),
(0.02, 100, 'right hand is to the UP of the left hand.')
]
]
self.palm_palm_relative_position_events = dict()
for i, axis in enumerate(['left-right', 'front-back', 'up-down']):
component = self.palm_palm_vec[:, i]
# print(f"axis: {axis} component:\n{component}")
events = split_events(
x=component,
x_intervals=relation_thre[i],
v_intervals=vel_thre,
delta_thre=0.05,
min_duration=4,
diff_state=True,
unit_len=self.unit_len,
)
if len(events) > 0:
events = [
{k: v for k, v in x.items() if k != 'direction'}
for x in events
]
self.palm_palm_relative_position_events[axis] = events
def extract_events(self):
self.lft_motioncode.extract_events()
self.rgt_motioncode.extract_events()
self.split_finger_finger_events()
self.split_finger_palm_events()
self.split_palm_palm_events()
def get_json(self):
events_summary = {
'frame_count': self.lft_motion.shape[0],
'left_hand_events': self.lft_motioncode.print_json(),
'right_hand_events': self.rgt_motioncode.print_json(),
'two_hand_relationships': {
'finger_tip_contact': self.finger_finger_distance_events,
'finger_palm_distance': self.finger_palm_distance_events,
'palm_palm_distance': self.palm_palm_distance_events,
'palm_palm_relative_position': self.palm_palm_relative_position_events
}
}
return events_summary
def generate_motion_codes(self):
self.extract_feats()
self.extract_events() |