-
Notifications
You must be signed in to change notification settings - Fork 10
/
Copy pathtest.py
executable file
·341 lines (310 loc) · 17.5 KB
/
test.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
import torch
import sys, os
sys.path.insert(0, os.path.dirname(__file__))
from LaFan import LaFan1
from torch.utils.data import Dataset, DataLoader
from model import StateEncoder, \
OffsetEncoder, \
TargetEncoder, \
LSTM, \
Decoder, \
ShortMotionDiscriminator, \
LongMotionDiscriminator
from skeleton import Skeleton
import torch.optim as optim
from tensorboardX import SummaryWriter
import numpy as np
from tqdm import tqdm
from functions import gen_ztta, write_to_bvhfile
import yaml
import time
import shutil
import imageio
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import axes3d, Axes3D
from remove_fs import remove_fs, save_bvh_from_network_output
from foot_sliding.animation_data import y_rotation_from_positions
from PIL import Image
def plot_pose(pose, cur_frame, prefix):
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
parents = [-1, 0, 1, 2, 3, 0, 5, 6, 7, 0, 9, 10, 11, 12, 11, 14, 15, 16, 11, 18, 19, 20]
ax.cla()
num_joint = pose.shape[0] // 3
for i, p in enumerate(parents):
if i > 0:
ax.plot([pose[i, 0], pose[p, 0]],\
[pose[i, 2], pose[p, 2]],\
[pose[i, 1], pose[p, 1]], c='r')
ax.plot([pose[i+num_joint, 0], pose[p+num_joint, 0]],\
[pose[i+num_joint, 2], pose[p+num_joint, 2]],\
[pose[i+num_joint, 1], pose[p+num_joint, 1]], c='b')
ax.plot([pose[i+num_joint*2, 0], pose[p+num_joint*2, 0]],\
[pose[i+num_joint*2, 2], pose[p+num_joint*2, 2]],\
[pose[i+num_joint*2, 1], pose[p+num_joint*2, 1]], c='g')
# ax.scatter(pose[:num_joint, 0], pose[:num_joint, 2], pose[:num_joint, 1],c='b')
# ax.scatter(pose[num_joint:num_joint*2, 0], pose[num_joint:num_joint*2, 2], pose[num_joint:num_joint*2, 1],c='b')
# ax.scatter(pose[num_joint*2:num_joint*3, 0], pose[num_joint*2:num_joint*3, 2], pose[num_joint*2:num_joint*3, 1],c='g')
xmin = np.min(pose[:, 0])
ymin = np.min(pose[:, 2])
zmin = np.min(pose[:, 1])
xmax = np.max(pose[:, 0])
ymax = np.max(pose[:, 2])
zmax = np.max(pose[:, 1])
scale = np.max([xmax - xmin, ymax - ymin, zmax - zmin])
xmid = (xmax + xmin) // 2
ymid = (ymax + ymin) // 2
zmid = (zmax + zmin) // 2
ax.set_xlim(xmid - scale // 2, xmid + scale // 2)
ax.set_ylim(ymid - scale // 2, ymid + scale // 2)
ax.set_zlim(zmid - scale // 2, zmid + scale // 2)
plt.draw()
plt.savefig(prefix + '_' + str(cur_frame)+'.png', dpi=200, bbox_inches='tight')
plt.close()
if __name__ == '__main__':
opt = yaml.load(open('./config/test-base.yaml', 'r').read())
model_dir =opt['test']['model_dir']
## initilize the skeleton ##
skeleton_mocap = Skeleton(offsets=opt['data']['offsets'], parents=opt['data']['parents'])
skeleton_mocap.cuda()
skeleton_mocap.remove_joints(opt['data']['joints_to_remove'])
## load train data ##
lafan_data_test = LaFan1(opt['data']['data_dir'], \
seq_len = opt['model']['seq_length'], \
offset = 40,\
train = False, debug=opt['test']['debug'])
lafan_data_test.cur_seq_length = opt['model']['seq_length']
x_mean = lafan_data_test.x_mean.cuda()
x_std = lafan_data_test.x_std.cuda().view(1, 1, opt['model']['num_joints'], 3)
lafan_loader_test = DataLoader(lafan_data_test, \
batch_size=opt['test']['batch_size'], \
shuffle=False, num_workers=opt['data']['num_workers'])
## initialize model and load parameters ##
state_encoder = StateEncoder(in_dim=opt['model']['state_input_dim'])
state_encoder = state_encoder.cuda()
state_encoder.load_state_dict(torch.load(os.path.join(opt['test']['model_dir'], 'state_encoder.pkl')))
offset_encoder = OffsetEncoder(in_dim=opt['model']['offset_input_dim'])
offset_encoder = offset_encoder.cuda()
offset_encoder.load_state_dict(torch.load(os.path.join(opt['test']['model_dir'], 'offset_encoder.pkl')))
target_encoder = TargetEncoder(in_dim=opt['model']['target_input_dim'])
target_encoder = target_encoder.cuda()
target_encoder.load_state_dict(torch.load(os.path.join(opt['test']['model_dir'], 'target_encoder.pkl')))
lstm = LSTM(in_dim=opt['model']['lstm_dim'], hidden_dim = opt['model']['lstm_dim'] * 2)
lstm = lstm.cuda()
lstm.load_state_dict(torch.load(os.path.join(opt['test']['model_dir'], 'lstm.pkl')))
decoder = Decoder(in_dim=opt['model']['lstm_dim'] * 2, out_dim=opt['model']['state_input_dim'])
decoder = decoder.cuda()
decoder.load_state_dict(torch.load(os.path.join(opt['test']['model_dir'], 'decoder.pkl')))
print('model loaded')
## get positional code ##
if opt['test']['use_ztta']:
ztta = gen_ztta().cuda()
# print('ztta:', ztta.size())
# assert 0
version = opt['test']['version']
# writer = SummaryWriter(log_dir)
loss_total_min = 10000000.0
for epoch in range(opt['test']['num_epoch']):
state_encoder.eval()
offset_encoder.eval()
target_encoder.eval()
lstm.eval()
decoder.eval()
loss_total_list = []
for i_batch, sampled_batch in enumerate(lafan_loader_test):
# if i_batch != 33:
# continue
pred_img_list = []
gt_img_list = []
img_list = []
# print(i_batch, sample_batched['local_q'].size())
loss_pos = 0
loss_quat = 0
loss_contact = 0
loss_root = 0
with torch.no_grad():
# if True:
# state input
local_q = sampled_batch['local_q'].cuda()
root_v = sampled_batch['root_v'].cuda()
contact = sampled_batch['contact'].cuda()
# offset input
root_p_offset = sampled_batch['root_p_offset'].cuda()
local_q_offset = sampled_batch['local_q_offset'].cuda()
local_q_offset = local_q_offset.view(local_q_offset.size(0), -1)
# target input
target = sampled_batch['target'].cuda()
target = target.view(target.size(0), -1)
# root pos
root_p = sampled_batch['root_p'].cuda()
# X
X = sampled_batch['X'].cuda()
bs = 6#np.random.choice(X.size(0), 1)[0]
if False:
print('local_q:', local_q.size(), \
'root_v:', root_v.size(), \
'contact:', contact.size(), \
'root_p_offset:', root_p_offset.size(), \
'local_q_offset:', local_q_offset.size(), \
'target:', target.size())
assert 0
lstm.init_hidden(local_q.size(0))
h_list = []
quat_list = []
quat_list.append(local_q[:,0,].view(local_q.size(0), -1, 4))
pred_list = []
pred_list.append(X[:,0])
bvh_list = []
bvh_list.append(torch.cat([X[:,0,0], local_q[:,0,].view(local_q.size(0), -1)], -1))
contact_list = []
contact_list.append(contact[:,0])
root_list = []
root_list.append(X[:,0,0])
# print(X.size())
for t in range(opt['model']['seq_length'] - 1):
# root pos
if t == 0:
root_p_t = root_p[:,t]
local_q_t = local_q[:,t]
local_q_t = local_q_t.view(local_q_t.size(0), -1)
contact_t = contact[:,t]
root_v_t = root_v[:,t]
else:
root_p_t = root_pred[0]
local_q_t = local_q_pred[0]
contact_t = contact_pred[0]
root_v_t = root_v_pred[0]
# state input
state_input = torch.cat([local_q_t, root_v_t, contact_t], -1)
# offset input
root_p_offset_t = root_p_offset - root_p_t
local_q_offset_t = local_q_offset - local_q_t
# print('root_p_offset_t:', root_p_offset_t.size(), 'local_q_offset_t:', local_q_offset_t.size())
offset_input = torch.cat([root_p_offset_t, local_q_offset_t], -1)
# target input
target_input = target
# print('state_input:',state_input.size())
h_state = state_encoder(state_input)
h_offset = offset_encoder(offset_input)
h_target = target_encoder(target_input)
if opt['test']['use_ztta']:
h_state += ztta[:, t]
h_offset += ztta[:, t]
h_target += ztta[:, t]
if opt['test']['use_adv']:
tta = opt['model']['seq_length'] - 2 - t
if tta < 5:
lambda_target = 0.0
elif tta >=5 and tta < 30:
lambda_target = (tta - 5) / 25.0
else:
lambda_target = 1.0
h_offset += 0.5 * lambda_target * torch.cuda.FloatTensor(h_offset.size()).normal_()
h_target += 0.5 * lambda_target * torch.cuda.FloatTensor(h_target.size()).normal_()
h_in = torch.cat([h_state, h_offset, h_target], -1).unsqueeze(0)
h_out = lstm(h_in)
# print('h_out:', h_out.size())
h_pred, contact_pred = decoder(h_out)
local_q_v_pred = h_pred[:,:,:opt['model']['target_input_dim']]
local_q_pred = local_q_v_pred + local_q_t
# print('q_pred:', q_pred.size())
local_q_pred_ = local_q_pred.view(local_q_pred.size(0), local_q_pred.size(1), -1, 4)
local_q_pred_ = local_q_pred_ / torch.norm(local_q_pred_, dim = -1, keepdim = True)
# print("local_q_pred_:", local_q_pred_.size())
quat_list.append(local_q_pred_[0])
root_v_pred = h_pred[:,:,opt['model']['target_input_dim']:]
root_pred = root_v_pred + root_p_t
root_list.append(root_pred[0])
# print(''contact:'', contact_pred.size())
# print('root_pred:', root_pred.size())
bvh_list.append(torch.cat([root_pred[0], local_q_pred_[0].view(local_q_pred_.size(1), -1)], -1))
pos_pred = skeleton_mocap.forward_kinematics(local_q_pred_, root_pred)
pos_next = X[:,t+1]
local_q_next = local_q[:,t+1]
local_q_next = local_q_next.view(local_q_next.size(0), -1)
root_p_next = root_p[:,t+1]
contact_next = contact[:,t+1]
# print(pos_pred.size(), x_std.size())
loss_pos += torch.mean(torch.abs(pos_pred[0] - pos_next) / x_std) / opt['model']['seq_length']
loss_quat += torch.mean(torch.abs(local_q_pred[0] - local_q_next)) / opt['model']['seq_length']
loss_root += torch.mean(torch.abs(root_pred[0] - root_p_next) / x_std[:,:,0]) / opt['model']['seq_length']
loss_contact += torch.mean(torch.abs(contact_pred[0] - contact_next)) / opt['model']['seq_length']
pred_list.append(pos_pred[0])
contact_list.append(contact_pred[0])
# if i_batch < 49:
# print("pos_pred:", pos_pred.size())
if opt['test']['save_img']:
plot_pose(np.concatenate([X[bs,0].view(22, 3).detach().cpu().numpy(),\
pos_pred[0, bs].view(22, 3).detach().cpu().numpy(),\
X[bs,-1].view(22, 3).detach().cpu().numpy()], 0),\
t, '../results'+version+'/pred')
plot_pose(np.concatenate([X[bs,0].view(22, 3).detach().cpu().numpy(),\
X[bs,t+1].view(22, 3).detach().cpu().numpy(),\
X[bs,-1].view(22, 3).detach().cpu().numpy()], 0),\
t, '../results'+version+'/gt')
pred_img = Image.open('../results'+version+'/pred_'+str(t)+'.png', 'r')
gt_img = Image.open('../results'+version+'/gt_'+str(t)+'.png', 'r')
pred_img_list.append(pred_img)
gt_img_list.append(gt_img)
img_list.append(np.concatenate([pred_img, gt_img.resize(pred_img.size)], 1))
# print('pivots:', pivots.shape)
# print('rot_data.size:', rot_data.shape)
if opt['test']['save_bvh']:
# print("bs:", bs)
bvh_data = torch.cat([x[bs].unsqueeze(0) for x in bvh_list], 0).detach().cpu().numpy()
# print('bvh_data:', bvh_data.shape)
# print('bvh_data:', bvh_data[0,3:7])
# assert 0
write_to_bvhfile(bvh_data, ('../bvh_seq/test_%03d.bvh' % i_batch), opt['data']['joints_to_remove'])
# assert 0
contact_data = torch.cat([x[bs].unsqueeze(0) for x in contact_list], 0).detach().cpu().numpy()
# rot_data = torch.cat([x[bs].unsqueeze(0) for x in quat_list], 0).detach().cpu().numpy()
# root_data = torch.cat([x[bs].unsqueeze(0) for x in root_list], 0).detach().cpu().numpy()
# pred_pose = torch.cat([x[bs].unsqueeze(0) for x in pred_list], 0).detach().cpu().numpy()
# quaters, pivots = y_rotation_from_positions(pred_pose, hips = (1,5), sdrs = (14,18))
# motion = np.concatenate([rot_data.reshape(rot_data.shape[0], -1),\
# root_data,\
# pivots], -1)
# motion = motion.transpose(1,0)
foot = contact_data.transpose(1,0)
foot[foot > 0.5] = 1.0
foot[foot <= 0.5] = 0.0
# print('foot[0]:',foot[0])
glb = remove_fs(('../bvh_seq/test_%03d.bvh' % i_batch), \
foot, \
fid_l=(3, 4), \
fid_r=(7, 8),\
output_path=("../bvh_seq_after"+version+"/test_%03d.bvh" % i_batch))
fix_img_list = []
for t in range(opt['model']['seq_length']):
plot_pose(np.concatenate([X[bs,0].view(22, 3).detach().cpu().numpy(),\
glb[t],\
X[bs,-1].view(22, 3).detach().cpu().numpy()], 0),\
t, '../results'+version+'/fixed')
plot_pose(np.concatenate([X[bs,0].view(22, 3).detach().cpu().numpy(),\
X[bs,t].view(22, 3).detach().cpu().numpy(),\
X[bs,-1].view(22, 3).detach().cpu().numpy()], 0),\
t, '../results'+version+'/gt')
fix_img = Image.open('../results'+version+'/fixed_'+str(t)+'.png', 'r')
gt_img = Image.open('../results'+version+'/gt_'+str(t)+'.png', 'r')
fix_img_list.append(np.concatenate([fix_img, gt_img.resize(fix_img.size)], 1))
imageio.mimsave(('../gif'+version+'/img_fix_%03d.gif' % i_batch), fix_img_list, duration=0.1)
# save_bvh_from_network_output(motion, output_path=("../bvh_seq_after/test_%03d.bvh" % i_batch))
# if i_batch < 49:
if opt['test']['save_img'] and opt['test']['save_gif']:
imageio.mimsave(('../gif'+version+'/img_%03d.gif' % i_batch), img_list, duration=0.1)
if opt['test']['save_pose']:
gt_pose = X[bs,:].view(opt['model']['seq_length'], 22, 3).detach().cpu().numpy()
pred_pose = torch.cat([x[bs].unsqueeze(0) for x in pred_list], 0).detach().cpu().numpy()
plt.clf()
joint_idx = 13
plt.plot(range(opt['model']['seq_length']), gt_pose[:,joint_idx,0])
plt.plot(range(opt['model']['seq_length']), pred_pose[:,joint_idx,0])
plt.legend(['gt', 'pred'])
plt.savefig('../results'+version+'/pose_%03d.png' % i_batch)
plt.close()
# if opt['test']['save_img'] and i_batch > 49:
# break
if opt['test']['save_pose'] and i_batch > 49:
break
# print("train epoch: %03d, cur total loss:%.3f, cur best loss:%.3f" % (epoch, loss_total_cur, loss_total_min))