-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataloader_integrate_act.py
408 lines (349 loc) · 17.5 KB
/
dataloader_integrate_act.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
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
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
import numpy as np
import sys
import os
import torch
import scipy.misc
import pickle
import matplotlib.pyplot as plt
import random
from mpl_toolkits.mplot3d import Axes3D
from torch.utils.data import Dataset
from interpolate import valid_crop_resize_multi_data
from numpy import inf
from collections import Counter
joint_seq = [(1, 2), (2, 21), (3, 21), (4, 3), (5, 21),
(6, 5), (7, 6), (8, 7), (9, 21), (10, 9),
(11, 10), (12, 11), (13, 1), (14, 13), (15, 14),
(16, 15), (17, 1), (18, 17), (19, 18), (20, 19),
(22, 23), (23, 8), (24, 25), (25, 12)]
def reshape_data_vacnn(data):
data = np.transpose(data,(1,3,2,0))
data = np.reshape(data,(data.shape[0],data.shape[1],data.shape[2]*data.shape[3]))
data = np.reshape(data,(data.shape[0],data.shape[1]*data.shape[2]))
data = np.expand_dims(data,axis=0)
return data
def calc_num_frames_vacnn(data,cvm):
data = reshape_data_vacnn(data)
zero_row = []
cvm = int(cvm)
ske_joint = np.squeeze(data,axis=0)
#print (ske_joint.shape)
for i in range(len(ske_joint)):
if (ske_joint[i, :] == np.zeros((1, cvm))).all():
zero_row.append(i)
ske_joint = np.delete(ske_joint, zero_row, axis=0)
return (ske_joint.shape[0])
def convert_to_rgb_vacnn(data,max_val,min_val,cvm):
zero_row = []
cvm = int(cvm)
ske_joint = np.squeeze(data,axis=0)
#print (ske_joint.shape)
for i in range(len(ske_joint)):
if (ske_joint[i, :] == np.zeros((1, cvm))).all():
zero_row.append(i)
ske_joint = np.delete(ske_joint, zero_row, axis=0)
if (ske_joint[:, 0:cvm//2] == np.zeros((ske_joint.shape[0], cvm//2))).all():
ske_joint = np.delete(ske_joint, range(cvm//2), axis=1)
elif (ske_joint[:, cvm//2:cvm] == np.zeros((ske_joint.shape[0], cvm//2))).all():
ske_joint = np.delete(ske_joint, range(cvm//2, cvm), axis=1)
# Convert to RGB
ske_joint = 255 * (ske_joint - min_val) / (max_val - min_val)
rgb_ske = np.reshape(ske_joint, (ske_joint.shape[0], ske_joint.shape[1] //3, 3))
if rgb_ske.shape[0]==0:
rgb_ske = np.zeros([224,224,3],dtype=np.float32)
else:
rgb_ske = scipy.misc.imresize(rgb_ske, (224, 224)).astype(np.float32)
#rgb_ske = np.array(Image.fromarray(rgb_ske.astype('uint8')).resize((224,224))).astype(np.float32)
rgb_ske = center(rgb_ske)
rgb_ske = np.transpose(rgb_ske, [1, 0, 2])
rgb_ske = np.transpose(rgb_ske, [2,1,0])
return rgb_ske
def convert_to_rgb_interpolate(data,max_val,min_val,cvm,center_flag):
zero_row = []
cvm = int(cvm)
ske_joint = np.squeeze(data,axis=0)
for i in range(len(ske_joint)):
if (ske_joint[i, :] == np.zeros((1, cvm))).all():
zero_row.append(i)
ske_joint = np.delete(ske_joint, zero_row, axis=0)
if (ske_joint[:, 0:cvm//2] == np.zeros((ske_joint.shape[0], cvm//2))).all():
ske_joint = np.delete(ske_joint, range(cvm//2), axis=1)
elif (ske_joint[:, cvm//2:cvm] == np.zeros((ske_joint.shape[0], cvm//2))).all():
ske_joint = np.delete(ske_joint, range(cvm//2, cvm), axis=1)
ske_joint = 255 * (ske_joint - min_val) / (max_val - min_val)
rgb_ske = np.reshape(ske_joint, (ske_joint.shape[0], ske_joint.shape[1] //3, 3))
rgb_ske = np.transpose(rgb_ske, [1, 0, 2])
rgb_ske = np.transpose(rgb_ske, [2,1,0])
return rgb_ske
def normalize_data(data,max_val,min_val,range_num):
if range_num == 0:
data = (data-min_val)/(max_val-min_val)
else:
data = 2*(data-min_val)/(max_val-min_val) - 1
return data
def vis_3dske(data_3d,num_frames_3d,save_dir,sample,normalize_flag,spherical_flag):
if spherical_flag == 1:
num_joints = data_3d.shape[2]
data_sph = np.zeros((3,num_frames_3d,num_joints),dtype=np.float32)
for i in range(num_frames_3d):
for j in range(num_joints):
az, el, r = data_3d[0,i,j], data_3d[1,i,j], data_3d[2,i,j]
#print (x,y,z)
x, y, z = sph2cart(az,el,r)
#print (az,el,r)
#print (x,y,z)
data_sph[0,:,:] = x
data_sph[1,:,:] = y
data_sph[2,:,:] = z
data_per1 = data_sph
else:
data_per1 = data_3d
data_per1 = np.expand_dims(data_per1,axis=3)
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for frame in range(num_frames_3d):
world_head_bones_per1_x = []
world_head_bones_per1_y = []
world_head_bones_per1_z = []
world_head_per1 = data_per1[:,frame,:,:]
world_head_per1 = np.squeeze(world_head_per1,axis=2)
world_head_per1 = np.transpose(world_head_per1,(1,0))
world_head_per1_x = world_head_per1[:,0]
world_head_per1_y = world_head_per1[:,1]
world_head_per1_z = world_head_per1[:,2]
# All bones
for i in range(len(joint_seq)):
#print (world_head_x[joint_seq[i][0]],world_head_x[joint_seq[i][1]])
world_head_bones_per1_x.append([world_head_per1_x[joint_seq[i][0]-1],world_head_per1_x[joint_seq[i][1]-1]])
world_head_bones_per1_y.append([world_head_per1_y[joint_seq[i][0]-1],world_head_per1_y[joint_seq[i][1]-1]])
world_head_bones_per1_z.append([world_head_per1_z[joint_seq[i][0]-1],world_head_per1_z[joint_seq[i][1]-1]])
world_head_bones_per1_x = np.array(world_head_bones_per1_x)
world_head_bones_per1_y = np.array(world_head_bones_per1_y)
world_head_bones_per1_z = np.array(world_head_bones_per1_z)
plt.cla()
for i in range(len(joint_seq)):
ax.plot(world_head_bones_per1_z[i],world_head_bones_per1_x[i],world_head_bones_per1_y[i],color='blue')
ax.scatter(world_head_per1_z,world_head_per1_x,world_head_per1_y,s=50,label='True Position')
if normalize_flag == 1:
ax.set_xlim(-0.5,0.5)
ax.set_ylim(-0.5,0.5)
ax.set_zlim(-0.5,0.5)
else:
ax.set_xlim(-1,4)
ax.set_ylim(-1,2)
ax.set_zlim(-1,2)
#if flag_rgb_to_3d == 1:
# ax2.set_xlim(0,1)
# ax2.set_ylim(0,1)
# ax2.set_zlim(0,1)
ax.set_title('Bones, frame={}'.format(frame))
plt.ioff()
plt.savefig(save_dir + "/sam%05dim%03d.png" % (sample,frame))
plt.close('all')
def center(rgb):
rgb[:,:,0] -= 110
rgb[:,:,1] -= 110
rgb[:,:,2] -= 110
return rgb
def cart2sph(x, y, z):
hxy = np.hypot(x, y)
r = np.hypot(hxy, z)
el = np.arctan2(z, hxy)
az = np.arctan2(y, x)
return az, el, r
def sph2cart(az, el, r):
rcos_theta = r * np.cos(el)
x = rcos_theta * np.cos(az)
y = rcos_theta * np.sin(az)
z = r * np.sin(el)
return x, y, z
class NTURGBDData1_act(Dataset):
def __init__(self,temporal_length,temporal_pattern,gpu_id,dataset,split,normalize_flag,centering_flag,spherical_flag,syn_per,only_syn,set_,expt):
self.temporal_length = temporal_length
self.labels = None
self.temporal_pattern = temporal_pattern
self.set_=set_
self.split = split
self.gpu_id = gpu_id
self.dataset = dataset
self.expt = expt
self.normalize = normalize_flag
self.centering = centering_flag
self.spherical = spherical_flag
self.syn_per = syn_per
self.only_syn = only_syn
if self.dataset == 'vacnn':
self.data_path_3d = "/fast-stripe/datasets/synthetic_data/ntu_3d_data_vacnn/"
self.deid_path = "/fast-stripe/datasets/synthetic_data/ntu_3d_data_vacnn/"
elif self.dataset == 'dgnn':
self.data_path_3d = "/fast-stripe/datasets/synthetic_data/ntu_3d_data_dgnn/"
self.deid_path = "/fast-stripe/datasets/synthetic_data/deid_multiperson_data/"
else:
self.data_path_3d = "/fast-stripe/datasets/synthetic_data/ntu_3d_data/"
self.deid_path = "/fast-stripe/datasets/synthetic_data/deid_multiperson_data/"
data_3d = set_+"_"+split+"_"+"3ddata.npy"
labels_3d = set_+"_"+split+"_"+"label_3ddata.pkl"
num_frames_3d = set_+"_"+split+"_"+"num_frame_3ddata.npy"
data_bone_3d = set_+"_"+split+"_"+"data_bone.npy"
self.data_bone_3d_path = os.path.join(self.data_path_3d,data_bone_3d)
self.data_3d_path = os.path.join(self.data_path_3d,data_3d)
self.labels_3d_path = os.path.join(self.data_path_3d,labels_3d)
self.num_frames_3d_path = os.path.join(self.data_path_3d,num_frames_3d)
with open(self.labels_3d_path,"rb") as f:
self.video_name,self.labels_3d = pickle.load(f)
f.close()
self.data_3d = np.load(self.data_3d_path)
self.num_frames_3d = np.load(self.num_frames_3d_path)
#============================================ Single-person Real ACTS ================================#
ignore_samples_path = '/fast-stripe/datasets/nyu_action/samples_with_missing_skeleton.txt'
with open(ignore_samples_path, 'r') as f:
ignored_samples = [line.strip() for line in f.readlines()]
f.close()
if self.temporal_pattern == 'interpolate':
if self.only_syn == 0:
tot_samples = 0
for i in range(len(self.labels_3d)):
if (self.labels_3d[i] <= 48): # <=48 : Single person acts ; 42 - Fall ; 26 - Jump
video_name = self.video_name[i].replace(".skeleton","")
if video_name not in ignored_samples:
tot_samples = tot_samples+1
#print ("Total single activity samples",tot_samples)
self.data_3d_inter = np.zeros((tot_samples,3,300,25,2),dtype=np.float32)
self.num_frames_3d_inter = np.zeros((tot_samples),dtype=np.int32)
self.video_name_inter = []
self.labels_3d_inter = []
self.type_label = [] # 1-Real data ; 0-Synthetic data
inter_ind = 0
for i in range(len(self.labels_3d)):
#print (self.labels_3d[i])
if (self.labels_3d[i] <= 48): # <=48 : Single person acts ; 42 - Fall ; 26 - Jump
video_name = self.video_name[i].replace(".skeleton","")
if video_name not in ignored_samples:
self.data_3d_inter[inter_ind] = self.data_3d[i]
#print (self.data_3d_inter[inter_ind])
self.num_frames_3d_inter[inter_ind] = self.num_frames_3d[i]
#print (self.num_frames_3d_inter)
self.video_name_inter.append(self.video_name[i])
self.labels_3d_inter.append(self.labels_3d[i])
self.type_label.append(1)
inter_ind = inter_ind + 1
self.data_3d = self.data_3d_inter
self.num_frames_3d = self.num_frames_3d_inter
self.video_name = self.video_name_inter
self.labels_3d = self.labels_3d_inter
#============================================ Single-person Synthetic ACTS ================================#
if self.set_ == 'train' and self.syn_per != 0:
syn_samples = 0
for class_num in range(44,49):
data_path = os.path.join('/fast-stripe/workspaces/deval/synthetic-data/wgan_gp/',str(class_num))
data_path = os.path.join(data_path,'img60_center_nonorm_per2_ep5k/models/1/eval_op','syn_data_%d.npy'%self.syn_per)
syn_data = np.load(data_path)
syn_samples = syn_samples+syn_data.shape[0]
if self.only_syn == 1:
real_samples = 0
else:
real_samples = self.data_3d.shape[0]
real_syn_samples = syn_samples + real_samples
# Initiate total data
self.data_3d_tot = np.zeros((real_syn_samples,3,300,25,2),dtype=np.float32)
self.num_frames_3d_tot = np.zeros((real_syn_samples),dtype=np.int32)
self.video_name_tot = []
self.labels_3d_tot = []
self.type_label_tot = [] # 1-Real data ; 0-Synthetic data
# Transfer real data
if self.only_syn == 0:
self.data_3d_tot[0:real_samples,:,:,:,:] = self.data_3d
self.num_frames_3d_tot[0:real_samples] = self.num_frames_3d
for i in range(real_samples):
self.video_name_tot.append(self.video_name[i])
self.labels_3d_tot.append(self.labels_3d[i])
self.type_label_tot.append(self.type_label[i])
# Transfer synthetic data
start_ind = real_samples
for class_num in range(44,49):
data_path = os.path.join('/fast-stripe/workspaces/deval/synthetic-data/wgan_gp/',str(class_num))
data_path = os.path.join(data_path,'img60_center_nonorm_per2_ep5k/models/1/eval_op','syn_data_%d.npy'%self.syn_per)
syn_data = np.load(data_path)
num_samples = syn_data.shape[0]
end_ind = start_ind + num_samples
self.data_3d_tot[start_ind:end_ind,:,0:self.temporal_length,:,:] = syn_data
for ind in range(start_ind,end_ind):
self.num_frames_3d_tot[ind] = self.temporal_length
self.video_name_tot.append("Synthetic.skeleton")
self.labels_3d_tot.append(class_num)
self.type_label_tot.append(0)
start_ind = end_ind
self.data_3d = self.data_3d_tot
self.video_name = self.video_name_tot
self.num_frames_3d = self.num_frames_3d_tot
self.labels_3d = self.labels_3d_tot
self.type_label = self.type_label_tot
#print (self.data_3d_inter.shape[0],self.num_frames_3d_inter.shape[0])
#print (self.data_3d.shape[0],self.num_frames_3d.shape[0],len(self.video_name))
print ("Loaded the %s set"%set_,"Total samples",len(self.video_name))
def __len__(self):
if (self.expt == 'check'):
return (len(self.video_name[0:10]))
else:
return (len(self.video_name))
def __getitem__(self,id):
# Get the labels
self.labels = self.labels_3d[id]
self.video = self.video_name[id]
self.num_frames = self.num_frames_3d[id]
self.type_ = self.type_label[id]
if self.dataset != 'vacnn':
self.video = self.video.replace(".skeleton","")
# Get the data
self.data = self.data_3d[id]
self.data = np.nan_to_num(self.data)
self.data[self.data == -inf] = 0
# Get the deid data & convert nans to 0.0
if self.temporal_pattern == 'interpolate':
if self.type_ == 1:
if self.centering == 1:
origin_data = self.data[:,:,1,0]
self.data = self.data - origin_data[:,:,None,None]
if self.set_ == 'train':
p_interval = [0.5,1]
p = np.random.rand(1)*(p_interval[1]-p_interval[0])+p_interval[0]
elif self.set_ == 'test':
p_interval = [0.95]
p = p_interval[0]
#print (self.data.shape,self.num_frames)
self.data = valid_crop_resize_multi_data(self.data,self.num_frames,p_interval,p,self.temporal_length)
#self.data = self.data[:,:,:,0:1]
#self.data = np.squeeze(self.data)
if self.normalize == 1:
#max_val,min_val = 5.18858098984,-5.28981208801
max_val,min_val = 5.826573,-4.9881773
self.data = normalize_data(self.data,max_val,min_val,-1)
if self.spherical == 1:
num_joints = self.data.shape[2]
data_sph = np.zeros((3,self.temporal_length,num_joints,2),dtype=np.float32)
for i in range(self.temporal_length):
for j in range(num_joints):
x,y,z = self.data[0,i,j,0], self.data[1,i,j,0], self.data[2,i,j,0]
#print (x,y,z)
az, el, r = cart2sph(x,y,z)
#print (az,el,r)
#print (x,y,z)
data_sph[0,:,:,0] = az
data_sph[1,:,:,0] = el
data_sph[1,:,:,0] = r
if self.centering == 1:
origin_data = data_sph[:,:,1,0]
data_sph = data_sph - origin_data[:,:,None,None]
self.data = data_sph
#print (self.data.shape)
#self.data = reshape_data_vacnn(self.data)
#rgb_ske = convert_to_rgb_interpolate(self.data,max_val,min_val,150,0)
#print (rgb_ske.shape)
elif self.type_ == 0:
self.data = self.data[:,0:self.temporal_length,:,:]
return self.data,self.labels
else:
max_val,min_val = 5.18858098984,-5.28981208801
self.data = reshape_data_vacnn(self.data)
rgb_ske = convert_to_rgb_vacnn(self.data,max_val,min_val,150) # C*V*M = 150
return rgb_ske,self.labels