-
Notifications
You must be signed in to change notification settings - Fork 1
/
lmdb_data_loader.py
310 lines (242 loc) · 12.1 KB
/
lmdb_data_loader.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
import datetime
import logging
import os
import pickle
import random
import numpy as np
import lmdb as lmdb
import torch
from torch.nn.utils.rnn import pad_sequence
from transformers import AutoTokenizer, AutoModel
from torch.utils.data import Dataset, DataLoader
from torch.utils.data.dataloader import default_collate
import utils.train_utils
import utils.data_utils
from model.vocab import Vocab
from data_loader.data_preprocessor import DataPreprocessor
import pyarrow
import librosa
import pdb
os.environ["TOKENIZERS_PARALLELISM"] = "false"
def word_seq_add_beat_collate_fn(data):
""" collate function for loading word sequences in variable lengths """
# sort a list by sequence length (descending order) to use pack_padded_sequence
data.sort(key=lambda x: len(x[0]), reverse=True)
# separate source and target sequences
beats, word_seq, text_padded, poses_seq, vec_seq, audio, spectrogram, aux_info = zip(*data)
# merge sequences
words_lengths = torch.LongTensor([len(x) for x in word_seq])
word_seq = pad_sequence(word_seq, batch_first=True).long()
beats = default_collate(beats)
text_padded = default_collate(text_padded)
poses_seq = default_collate(poses_seq)
vec_seq = default_collate(vec_seq)
audio = default_collate(audio)
spectrogram = default_collate(spectrogram)
aux_info = {key: default_collate([d[key] for d in aux_info]) for key in aux_info[0]}
return beats, word_seq, words_lengths, text_padded, poses_seq, vec_seq, audio, spectrogram, aux_info
def word_seq_collate_fn(data):
""" collate function for loading word sequences in variable lengths """
# sort a list by sequence length (descending order) to use pack_padded_sequence
data.sort(key=lambda x: len(x[0]), reverse=True)
# separate source and target sequences
_, word_seq, text_padded, poses_seq, vec_seq, audio, spectrogram, aux_info = zip(*data)
# merge sequences
words_lengths = torch.LongTensor([len(x) for x in word_seq])
word_seq = pad_sequence(word_seq, batch_first=True).long()
text_padded = default_collate(text_padded)
poses_seq = default_collate(poses_seq)
vec_seq = default_collate(vec_seq)
audio = default_collate(audio)
spectrogram = default_collate(spectrogram)
aux_info = {key: default_collate([d[key] for d in aux_info]) for key in aux_info[0]}
return word_seq, words_lengths, text_padded, poses_seq, vec_seq, audio, spectrogram, aux_info
def default_collate_fn(data):
_, text_padded, pose_seq, vec_seq, audio, spectrogram, aux_info = zip(*data)
text_padded = default_collate(text_padded)
pose_seq = default_collate(pose_seq)
vec_seq = default_collate(vec_seq)
audio = default_collate(audio)
spectrogram = default_collate(spectrogram)
aux_info = {key: default_collate([d[key] for d in aux_info]) for key in aux_info[0]}
return torch.tensor([0]), torch.tensor([0]), text_padded, pose_seq, vec_seq, audio, spectrogram, aux_info
class SpeechMotionDataset(Dataset):
def __init__(self, lmdb_dir, n_poses, subdivision_stride, pose_resampling_fps, mean_pose, mean_dir_vec,
speaker_model=None, remove_word_timing=False, save_flag=False):
self.lmdb_dir = lmdb_dir
self.n_poses = n_poses
self.subdivision_stride = subdivision_stride
self.skeleton_resampling_fps = pose_resampling_fps
self.mean_dir_vec = mean_dir_vec
self.remove_word_timing = remove_word_timing
self.expected_audio_length = int(round(n_poses / pose_resampling_fps * 16000))
self.expected_spectrogram_length = utils.data_utils.calc_spectrogram_length_from_motion_length(
n_poses, pose_resampling_fps)
self.lang_model = None
self.save_flag = save_flag
#self.beat_path = 'beat_resave'
self.beat_path = '../Gesture-Generation-from-Trimodal-Context/double_feat'
logging.info("Reading data '{}'...".format(lmdb_dir))
preloaded_dir = lmdb_dir + '_cache'
if not os.path.exists(preloaded_dir):
logging.info('Creating the dataset cache...')
assert mean_dir_vec is not None
if mean_dir_vec.shape[-1] != 3:
mean_dir_vec = mean_dir_vec.reshape(mean_dir_vec.shape[:-1] + (-1, 3))
n_poses_extended = int(round(n_poses * 1.25)) # some margin
data_sampler = DataPreprocessor(lmdb_dir, preloaded_dir, n_poses_extended,
subdivision_stride, pose_resampling_fps, mean_pose, mean_dir_vec)
data_sampler.run()
else:
logging.info('Found the cache {}'.format(preloaded_dir))
# init lmdb
self.lmdb_env = lmdb.open(preloaded_dir, readonly=True, lock=False)
with self.lmdb_env.begin() as txn:
self.n_samples = txn.stat()['entries']
# make a speaker model
if speaker_model is None or speaker_model == 0:
precomputed_model = lmdb_dir + '_speaker_model.pkl'
if not os.path.exists(precomputed_model):
self._make_speaker_model(lmdb_dir, precomputed_model)
else:
with open(precomputed_model, 'rb') as f:
self.speaker_model = pickle.load(f)
else:
self.speaker_model = speaker_model
def __len__(self):
return self.n_samples
def proc_audio(self, audio, audio_numpy):
audio_flag = (audio > 1e-3).float()
audio_thresh = (audio * audio_flag).sum() / audio_flag.sum()
audio_len = 340*3
ind = torch.arange(audio_len)
au_size = audio.size(0)
audio_step = au_size // audio_len
ind_step = ind * audio_step
sample_bias = audio_step // 2
ind_step_bias = ind_step + sample_bias
ind_step_bias[-1] = au_size - 1 if ind_step_bias[-1] >= au_size else ind_step_bias[-1]
audio_sample = audio.abs()[ind_step_bias]
audio_flag = (audio_sample > audio_thresh).float()
hop_len = au_size // audio_len
sr_num = 16000
oenv = librosa.onset.onset_strength(y=audio_numpy, sr=sr_num, hop_length=hop_len)
start_skip = (oenv.size - audio_len) // 2
audio_oenv = torch.from_numpy(oenv)[start_skip:audio_len+start_skip]
audio_beat = torch.stack([audio_flag, audio_oenv])
return audio_beat
def __getitem__(self, idx):
with self.lmdb_env.begin(write=False) as txn:
key = '{:010}'.format(idx).encode('ascii')
sample = txn.get(key)
sample = pyarrow.deserialize(sample)
word_seq, pose_seq, vec_seq, audio, spectrogram, aux_info = sample
def extend_word_seq(lang, words, end_time=None):
n_frames = self.n_poses
if end_time is None:
end_time = aux_info['end_time']
frame_duration = (end_time - aux_info['start_time']) / n_frames
extended_word_indices = np.zeros(n_frames) # zero is the index of padding token
if self.remove_word_timing:
n_words = 0
for word in words:
idx = max(0, int(np.floor((word[1] - aux_info['start_time']) / frame_duration)))
if idx < n_frames:
n_words += 1
space = int(n_frames / (n_words + 1))
for i in range(n_words):
idx = (i+1) * space
extended_word_indices[idx] = lang.get_word_index(words[i][0])
else:
prev_idx = 0
for word in words:
idx = max(0, int(np.floor((word[1] - aux_info['start_time']) / frame_duration)))
if idx < n_frames:
extended_word_indices[idx] = lang.get_word_index(word[0])
# extended_word_indices[prev_idx:idx+1] = lang.get_word_index(word[0])
prev_idx = idx
return torch.Tensor(extended_word_indices).long()
def words_to_tensor(lang, words, end_time=None):
indexes = [lang.SOS_token]
for word in words:
if end_time is not None and word[1] > end_time:
break
indexes.append(lang.get_word_index(word[0]))
indexes.append(lang.EOS_token)
return torch.Tensor(indexes).long()
duration = aux_info['end_time'] - aux_info['start_time']
do_clipping = True
if do_clipping:
sample_end_time = aux_info['start_time'] + duration * self.n_poses / vec_seq.shape[0]
audio = utils.data_utils.make_audio_fixed_length(audio, self.expected_audio_length)
spectrogram = spectrogram[:, 0:self.expected_spectrogram_length]
vec_seq = vec_seq[0:self.n_poses]
pose_seq = pose_seq[0:self.n_poses]
else:
sample_end_time = None
#word_list = []
word_list = [x[0] for x in word_seq]
start_time_list = [str(x[1]) for x in word_seq]
end_time_list = [str(x[2]) for x in word_seq]
sent = ' '.join(word_list)
rec_sent = '|'.join(word_list)
rec_start = '|'.join(start_time_list)
rec_end = '|'.join(end_time_list)
word_raw = rec_sent + ' ' + rec_start + ' ' + rec_end
aux_info['word_list'] = sent
aux_info['word_raw'] = word_raw
# to tensors
word_seq_tensor = words_to_tensor(self.lang_model, word_seq, sample_end_time)
extended_word_seq = extend_word_seq(self.lang_model, word_seq, sample_end_time)
vec_seq_ = torch.from_numpy(vec_seq.copy())
vec_seq_ = vec_seq_.reshape((vec_seq_.shape[0], -1)).float()
pose_seq = torch.from_numpy(pose_seq.copy()).reshape((pose_seq.shape[0], -1)).float()
audio_numpy = audio.copy().astype(np.float32)
audio = torch.from_numpy(audio.copy()).float()
aux_key = '_'.join([aux_info['vid'],
str(aux_info['start_frame_no']),
str(aux_info['end_frame_no'])])
if os.path.exists(self.beat_path):
audio_beat = torch.load(os.path.join(self.beat_path,
aux_key + '.pt'))
else:
audio_beat = self.proc_audio(audio, audio_numpy)
torch.save(audio_beat, os.path.join(self.beat_path, aux_key + '.pt'))
#sr_num = 16000
#duration = librosa.get_duration(y=audio_numpy, sr=sr_num)
#try:
# tempo, beats = librosa.beat.beat_track(y=audio_numpy, sr=sr_num)
# time_seq = librosa.frames_to_time(beats, sr=sr_num)
#except:
# time_seq = None
#seq_len_ = 340
#time_seq_tensor = torch.zeros(seq_len_)
#if time_seq is not None:
# time_seq_round = (seq_len_ * (time_seq / duration)).round()
# for cnt_, ind in enumerate(time_seq_round):
# ind = int(ind)
# time_seq_tensor[ind:] = time_seq_tensor[ind:] + (-1)**cnt_
#print(time_seq_round)
if self.save_flag:
torch.save(audio_beat,
os.path.join(self.beat_path, aux_key + '.pt') )
spectrogram = torch.from_numpy(spectrogram.copy())
return audio_beat, word_seq_tensor, extended_word_seq, pose_seq, vec_seq_, audio, spectrogram, aux_info
def set_lang_model(self, lang_model):
self.lang_model = lang_model
def _make_speaker_model(self, lmdb_dir, cache_path):
logging.info(' building a speaker model...')
speaker_model = Vocab('vid', insert_default_tokens=False)
lmdb_env = lmdb.open(lmdb_dir, readonly=True, lock=False)
txn = lmdb_env.begin(write=False)
cursor = txn.cursor()
for key, value in cursor:
video = pyarrow.deserialize(value)
vid = video['vid']
speaker_model.index_word(vid)
lmdb_env.close()
logging.info(' indexed %d videos' % speaker_model.n_words)
self.speaker_model = speaker_model
# cache
with open(cache_path, 'wb') as f:
pickle.dump(self.speaker_model, f)