-
Notifications
You must be signed in to change notification settings - Fork 41
/
Copy pathgen_motion_script.py
197 lines (157 loc) · 8.42 KB
/
gen_motion_script.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
import os
from os.path import join as pjoin
import utils.paramUtil as paramUtil
from options.evaluate_options import TestOptions
from torch.utils.data import DataLoader
from utils.plot_script import *
from networks.modules import *
from networks.trainers import CompTrainerV6
from data.dataset import RawTextDataset
from scripts.motion_process import *
from utils.word_vectorizer import WordVectorizer, POS_enumerator
from utils.utils import *
def plot_t2m(data, save_dir, captions):
data = dataset.inv_transform(data)
# print(ep_curves.shape)
for i, (caption, joint_data) in enumerate(zip(captions, data)):
joint = recover_from_ric(torch.from_numpy(joint_data).float(), opt.joints_num).numpy()
save_path = '%s_%02d'%(save_dir, i)
# np.save(save_path + '.npy', joint)
# plot_3d_motion(save_path + '.mp4', paramUtil.t2m_kinematic_chain, joint, title=caption, fps=20)
joint = motion_temporal_filter(joint, sigma=1)
np.save(save_path + '_a.npy', joint)
plot_3d_motion(save_path + '_a.mp4', paramUtil.t2m_kinematic_chain, joint, title=caption, fps=20)
def loadDecompModel(opt):
movement_enc = MovementConvEncoder(dim_pose - 4, opt.dim_movement_enc_hidden, opt.dim_movement_latent)
movement_dec = MovementConvDecoder(opt.dim_movement_latent, opt.dim_movement_dec_hidden, dim_pose)
checkpoint = torch.load(pjoin(opt.checkpoints_dir, opt.dataset_name, opt.decomp_name, 'model', 'latest.tar'),
map_location=opt.device)
movement_enc.load_state_dict(checkpoint['movement_enc'])
return movement_enc, movement_dec
def build_models(opt):
if opt.text_enc_mod == 'bigru':
text_encoder = TextEncoderBiGRU(word_size=dim_word,
pos_size=dim_pos_ohot,
hidden_size=opt.dim_text_hidden,
device=opt.device)
text_size = opt.dim_text_hidden * 2
else:
raise Exception("Text Encoder Mode not Recognized!!!")
seq_prior = TextDecoder(text_size=text_size,
input_size=opt.dim_att_vec + opt.dim_movement_latent,
output_size=opt.dim_z,
hidden_size=opt.dim_pri_hidden,
n_layers=opt.n_layers_pri)
seq_decoder = TextVAEDecoder(text_size=text_size,
input_size=opt.dim_att_vec + opt.dim_z + opt.dim_movement_latent,
output_size=opt.dim_movement_latent,
hidden_size=opt.dim_dec_hidden,
n_layers=opt.n_layers_dec)
att_layer = AttLayer(query_dim=opt.dim_pos_hidden,
key_dim=text_size,
value_dim=opt.dim_att_vec)
movement_enc = MovementConvEncoder(dim_pose - 4, opt.dim_movement_enc_hidden, opt.dim_movement_latent)
movement_dec = MovementConvDecoder(opt.dim_movement_latent, opt.dim_movement_dec_hidden, dim_pose)
# latent_dis = LatentDis(input_size=opt.dim_z * 2)
# return text_encoder, text_decoder, att_layer, vae_pri, vae_dec, vae_pos, motion_dis, movement_dis, latent_dis
return text_encoder, seq_prior, seq_decoder, att_layer, movement_enc, movement_dec
if __name__ == '__main__':
parser = TestOptions()
opt = parser.parse()
opt.do_denoise = True
opt.device = torch.device("cpu" if opt.gpu_id==-1 else "cuda:" + str(opt.gpu_id))
torch.autograd.set_detect_anomaly(True)
opt.save_root = pjoin(opt.checkpoints_dir, opt.dataset_name, opt.name)
opt.model_dir = pjoin(opt.save_root, 'model')
opt.meta_dir = pjoin(opt.save_root, 'meta')
opt.result_dir = pjoin(opt.result_path, opt.dataset_name, opt.name, opt.ext)
opt.joint_dir = pjoin(opt.result_dir, 'joints')
opt.animation_dir = pjoin(opt.result_dir, 'animations')
os.makedirs(opt.joint_dir, exist_ok=True)
os.makedirs(opt.animation_dir, exist_ok=True)
if opt.dataset_name == 't2m':
opt.data_root = './dataset/HumanML3D'
opt.motion_dir = pjoin(opt.data_root, 'new_joint_vecs')
opt.text_dir = pjoin(opt.data_root, 'texts')
opt.joints_num = 22
dim_pose = 263
dim_word = 300
dim_pos_ohot = len(POS_enumerator)
num_classes = 200 // opt.unit_length
mean = np.load(pjoin(opt.meta_dir, 'mean.npy'))
std = np.load(pjoin(opt.meta_dir, 'std.npy'))
w_vectorizer = WordVectorizer('./glove', 'our_vab')
split_file = pjoin(opt.data_root, opt.split_file)
opt.max_motion_length = 196
else:
raise KeyError('Dataset Does Not Exist')
text_enc, seq_pri, seq_dec, att_layer, mov_enc, mov_dec = build_models(opt)
# mov_enc, mov_dec = loadDecompModel(opt)
trainer = CompTrainerV6(opt, text_enc, seq_pri, seq_dec, att_layer, mov_dec, mov_enc=mov_enc)
# dataset = Text2MotionDataset(opt, mean, std, split_file, w_vectorizer)
dataset = RawTextDataset(opt, mean, std, opt.text_file, w_vectorizer)
# dataset.reset_max_len(opt.start_mov_len * opt.unit_length)
epoch, it, sub_ep, schedule_len = trainer.load(pjoin(opt.model_dir, opt.which_epoch + '.tar'))
print('Loading model: Epoch %03d Schedule_len %03d'%(epoch, schedule_len))
trainer.eval_mode()
trainer.to(opt.device)
# mov_enc.to(opt.device)
# mov2_dec.to(opt.device)
# if opt.est_length:
estimator = MotionLenEstimatorBiGRU(dim_word, dim_pos_ohot, 512, num_classes)
checkpoints = torch.load(pjoin(opt.checkpoints_dir, opt.dataset_name, 'length_est_bigru', 'model', 'latest.tar'), map_location=torch.device("cuda"))
estimator.load_state_dict(checkpoints['estimator'])
estimator.to(opt.device)
estimator.eval()
data_loader = DataLoader(dataset, batch_size=1, drop_last=True, num_workers=1)
'''Generate Results'''
print('Generate Results')
result_dict = {}
with torch.no_grad():
for i, data in enumerate(data_loader):
print('%02d_%03d'%(i, len(data_loader)))
word_emb, pos_ohot, caption, cap_lens = data
name = 'C%03d'%(i)
item_dict = {'caption': caption}
print(caption)
word_emb, pos_ohot, caption, cap_lens = data
word_emb = word_emb.detach().to(opt.device).float()
pos_ohot = pos_ohot.detach().to(opt.device).float()
pred_dis = estimator(word_emb, pos_ohot, cap_lens)
pred_dis = nn.Softmax(-1)(pred_dis).squeeze()
# pred_dis_np = pred_dis.cpu().numpy()
# max_idxs = pred_dis_np.argsort()[-5:][::-1]
# max_values = pred_dis_np[max_idxs]
# print(max_idxs)
# print(max_values)
# print(m_lens[0] // opt.unit_length)
for t in range(opt.repeat_times):
length = torch.multinomial(pred_dis, 1)
# print(length.item())
m_lens = length * opt.unit_length
pred_motions, _, att_wgts = trainer.generate(word_emb, pos_ohot, cap_lens, m_lens, m_lens[0]//opt.unit_length, dim_pose)
# trainer.forward(data, 0, m_lens[0]//opt.unit_length)
# pred_motions = trainer.pred_motions.view(opt.batch_size, m_lens[0], -1)
# ep_curves = trainer.ep_curve
sub_dict = {}
sub_dict['motion'] = pred_motions.cpu().numpy()
sub_dict['att_wgts'] = att_wgts.cpu().numpy()
sub_dict['m_len'] = m_lens[0]
item_dict['result_%02d'%t] = sub_dict
result_dict[name] = item_dict
print('Animation Results')
'''Animate Results'''
for i, (key, item) in enumerate(result_dict.items()):
print('%02d_%03d'%(i, len(result_dict)))
captions = item['caption']
joint_save_path = pjoin(opt.joint_dir, key)
animation_save_path = pjoin(opt.animation_dir, key)
os.makedirs(joint_save_path, exist_ok=True)
os.makedirs(animation_save_path, exist_ok=True)
for t in range(opt.repeat_times):
sub_dict = item['result_%02d'%t]
motion = sub_dict['motion']
att_wgts = sub_dict['att_wgts']
np.save(pjoin(joint_save_path, 'gen_motion_%02d_L%03d.npy' % (t, motion.shape[1])), motion)
# np.save(pjoin(joint_save_path, 'att_wgt_%02d_L%03d.npy' % (t, motion.shape[1])), att_wgts)
plot_t2m(motion, pjoin(animation_save_path, 'gen_motion_%02d_L%03d' % (t, motion.shape[1])), captions)