-
Notifications
You must be signed in to change notification settings - Fork 1
/
main.py
executable file
·534 lines (414 loc) · 18.6 KB
/
main.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
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
import os
import time
import torch
import numpy as np
import random
import scipy.io.wavfile as wavfile
import librosa
from arguments import ArgParser
from dataset import MUSICWaveDataset, MUSICSpecDataset
from models import ModelBuilder
from utils import AverageMeter, save_points, makedirs, center_trim
from viz import plot_loss_metrics
from metrics_binaural import compute_metrics
import MinkowskiEngine as ME
# Network wrapper, defines forward pass
class NetWrapper(torch.nn.Module):
def __init__(self, nets, crit):
super(NetWrapper, self).__init__()
self.net_sound, self.net_vision = nets
self.crit = crit
def forward(self, batch_data, args):
if args.arch_sound == 'demucs':
in_padded_audio = batch_data['in_padded_audio']
in_audio_shape = torch.zeros(in_padded_audio.shape[0], in_padded_audio.shape[1], args.audLen)
gt_audio = batch_data['out_audio']
feats = batch_data['feats']
coords = batch_data['coords']
# 2. forward net_vision
sin = ME.SparseTensor(feats, coords.int(), allow_duplicate_coords=True) #Create SparseTensor
visual_feature = self.net_vision.forward(sin)
# 3. forward audio_vision
pred_audio = self.net_sound.forward(in_padded_audio, visual_feature)
pred_audio = torch.squeeze(pred_audio, dim=2)
pred_audio = center_trim(pred_audio, in_audio_shape)
# 4. loss
err = self.crit(pred_audio, gt_audio)
return err, \
{'pred_audio': pred_audio}
elif args.arch_sound == 'unet':
audio_mix_spec = batch_data['audio_mix_spec']
audio_diff_spec = batch_data['audio_diff_spec']
audio_gt = audio_diff_spec[:, :, :-1, :]
feats = batch_data['feats']
coords = batch_data['coords']
# 2. forward net_vision
sin = ME.SparseTensor(feats, coords.int(), allow_duplicate_coords=True) #Create SparseTensor
visual_feature = self.net_vision.forward(sin)
# 3. forward audio_vision
pred_mask = self.net_sound.forward(audio_mix_spec, visual_feature)
# complex masking to obtain the predicted spectrogram
spectrogram_diff_real = audio_mix_spec[:, 0, :-1, :] * pred_mask[:, 0, :, :] - \
audio_mix_spec[:, 1, :-1, :] * pred_mask[:, 1, :, :]
spectrogram_diff_img = audio_mix_spec[:, 0, :-1, :] * pred_mask[:, 1, :, :] + \
audio_mix_spec[:, 1, :-1, :] * pred_mask[:, 0, :, :]
pred_diff_audio_spec = torch.cat((spectrogram_diff_real.unsqueeze(1), spectrogram_diff_img.unsqueeze(1)), 1)
# 4. loss
err = self.crit(pred_diff_audio_spec, audio_gt)
return err, \
{'pred_audio': pred_diff_audio_spec}
# Calculate metrics
def calc_metrics_waveform(batch_data, outputs):
# meters
envelope_meter = AverageMeter()
stft_l2_meter = AverageMeter()
# fetch data and predictions
gt_audio = batch_data['out_audio']
pred_audio = outputs['pred_audio']
# convert into numpy
gt_audio = gt_audio.detach().cpu().numpy()
pred_audio = pred_audio.detach().cpu().numpy()
# loop over each sample
B = gt_audio.shape[0]
for j in range(B):
gt_audio_j = gt_audio[j, ...]
pred_audio_j = pred_audio[j, ...]
# binaural performance computes
stft_l2, envelope_distance = compute_metrics(pred_audio_j, gt_audio_j)
stft_l2_meter.update(stft_l2)
envelope_meter.update(envelope_distance)
return [envelope_meter.average(),
stft_l2_meter.average()]
def calc_metrics_spectrogram(batch_data, outputs, args):
# meters
envelope_meter = AverageMeter()
stft_l2_meter = AverageMeter()
# fetch data and predictions
gt_audio = batch_data['out_audio']
pred_diff_audio_spec = outputs['pred_audio']
audio_mix = batch_data['in_audio']
# convert into numpy
gt_audio = gt_audio.detach().cpu().numpy()
pred_diff_audio_spec = pred_diff_audio_spec.detach().cpu().numpy()
audio_mix = audio_mix.detach().cpu().numpy()
# loop over each sample
B = gt_audio.shape[0]
for j in range(B):
gt_audio_j = gt_audio[j, ...]
pred_diff_audio_spec_j = pred_diff_audio_spec[j, ...]
audio_mix_j= audio_mix[j, ...]
# ISTFT to convert back to audio
reconstructed_stft_diff = pred_diff_audio_spec_j[0, :, :] + (1j * pred_diff_audio_spec_j[1, :, :])
reconstructed_signal_diff = librosa.istft(reconstructed_stft_diff, hop_length=441, win_length=1014, center=True,
length=args.audLen)
reconstructed_signal_left = (audio_mix_j + reconstructed_signal_diff) / 2
reconstructed_signal_right = (audio_mix_j - reconstructed_signal_diff) / 2
reconstructed_binaural = np.concatenate((reconstructed_signal_left, reconstructed_signal_right), axis=0)
# binaural performance computes
stft_l2, envelope_distance = compute_metrics(reconstructed_binaural, gt_audio_j)
stft_l2_meter.update(stft_l2)
envelope_meter.update(envelope_distance)
return [envelope_meter.average(),
stft_l2_meter.average()]
# Visualize predictions
def output_visuals_waveform(batch_data, outputs, args):
in_audio = batch_data['in_audio']
in_padded_audio = batch_data['in_padded_audio']
gt_audio = batch_data['out_audio']
features = batch_data['feats']
coords = batch_data['coords']
prefix = batch_data['prefix']
pred_audio = outputs['pred_audio']
# convert into numpy
in_audio = in_audio.detach().cpu().numpy()
in_padded_audio = in_padded_audio.detach().cpu().numpy()
gt_audio = gt_audio.detach().cpu().numpy()
pred_audio = pred_audio.detach().cpu().numpy()
in_audio = np.transpose(in_audio, (0, 2, 1))
in_padded_audio = np.transpose(in_padded_audio, (0, 2, 1))
gt_audio = np.transpose(gt_audio, (0, 2, 1))
pred_audio = np.transpose(pred_audio, (0, 2, 1))
B = in_audio.shape[0]
# loop over each sample
for j in range(B):
makedirs(os.path.join(args.vis, prefix[j]))
# save mono
filename_in_audio = os.path.join(prefix[j], 'in.wav')
wavfile.write(os.path.join(args.vis, filename_in_audio), args.audRate, in_audio[j, ...])
filename_in_audio = os.path.join(prefix[j], 'in_padded.wav')
wavfile.write(os.path.join(args.vis, filename_in_audio), args.audRate, in_padded_audio[j, ...])
# save output binaural
filename_gtwav = os.path.join(prefix[j], 'gt.wav')
filename_predwav = os.path.join(prefix[j], 'pred.wav')
wavfile.write(os.path.join(args.vis, filename_gtwav), args.audRate, gt_audio[j, ...])
wavfile.write(os.path.join(args.vis, filename_predwav), args.audRate, pred_audio[j, ...])
idx = torch.where(coords[:, 0] == j)
path_point = os.path.join(args.vis, prefix[j], 'point_cloud_scene.ply')
if args.rgbs_feature:
colors = np.asarray(features[idx])
xyz = np.asarray(coords[idx][:, 1:4])
xyz = xyz*args.voxel_size
save_points(path_point, xyz, colors)
else:
xyz = np.asarray(features[idx])
save_points(path_point, xyz)
# Visualize predictions
def output_visuals_spectrogram(batch_data, outputs, args):
in_audio = batch_data['in_audio']
audio_mix = in_audio
pred_diff_audio_spec = outputs['pred_audio']
gt_audio = batch_data['out_audio']
features = batch_data['feats']
coords = batch_data['coords']
prefix = batch_data['prefix']
# convert into numpy
in_audio = in_audio.detach().cpu().numpy()
gt_audio = gt_audio.detach().cpu().numpy()
pred_diff_audio_spec = pred_diff_audio_spec.detach().cpu().numpy()
in_audio = np.transpose(in_audio, (0, 2, 1))
gt_audio = np.transpose(gt_audio, (0, 2, 1))
B = in_audio.shape[0]
# loop over each sample
for j in range(B):
makedirs(os.path.join(args.vis, prefix[j]))
# save mono
filename_in_audio = os.path.join(prefix[j], 'in.wav')
wavfile.write(os.path.join(args.vis, filename_in_audio), args.audRate, in_audio[j, ...])
# save output binaural
filename_gtwav = os.path.join(prefix[j], 'gt.wav')
wavfile.write(os.path.join(args.vis, filename_gtwav), args.audRate, gt_audio[j, ...])
#ISTFT to convert pred back to audio
pred_diff_audio_spec_j = pred_diff_audio_spec[j, ...]
reconstructed_stft_diff = pred_diff_audio_spec_j[0, :, :] + (1j * pred_diff_audio_spec_j[1, :, :])
reconstructed_signal_diff = librosa.istft(reconstructed_stft_diff, hop_length=441, win_length=1014, center=True,
length=args.audLen)
reconstructed_signal_left = (audio_mix[j, ...] + reconstructed_signal_diff) / 2
reconstructed_signal_right = (audio_mix[j, ...] - reconstructed_signal_diff) / 2
pred_audio = np.concatenate((reconstructed_signal_left, reconstructed_signal_right), axis=0)
pred_audio = np.transpose(pred_audio, (1, 0))
filename_predwav = os.path.join(prefix[j], 'pred.wav')
wavfile.write(os.path.join(args.vis, filename_predwav), args.audRate, pred_audio)
idx = torch.where(coords[:, 0] == j)
path_point = os.path.join(args.vis, prefix[j], 'point_cloud_scene.ply')
if args.rgbs_feature:
colors = np.asarray(features[idx])
xyz = np.asarray(coords[idx][:, 1:4])
xyz = xyz*args.voxel_size
save_points(path_point, xyz, colors)
else:
xyz = np.asarray(features[idx])
save_points(path_point, xyz)
def evaluate(netWrapper, loader, history, epoch, args):
print('Evaluating at {} epochs...'.format(epoch))
torch.set_grad_enabled(False)
# remove previous viz results
makedirs(args.vis, remove=True)
# switch to eval mode
netWrapper.eval()
# initialize meters
loss_meter = AverageMeter()
envelope_meter = AverageMeter()
stft_l2_meter = AverageMeter()
for i, batch_data in enumerate(loader):
# forward pass
err, outputs = netWrapper.forward(batch_data, args)
loss_meter.update(err.item())
print('[Eval] iter {}, loss: {:.4f}'.format(i, err.item()))
if args.arch_sound == 'demucs':
# calculate metrics
envelope_distance, stft_l2 = calc_metrics_waveform(batch_data, outputs)
# output visualization
output_visuals_waveform(batch_data, outputs, args)
if args.arch_sound == 'unet':
# calculate metrics
envelope_distance, stft_l2 = calc_metrics_spectrogram(batch_data, outputs, args)
# output visualization
output_visuals_spectrogram(batch_data, outputs, args)
envelope_meter.update(envelope_distance)
stft_l2_meter.update(stft_l2)
print('[Eval Summary] Epoch: {}, Loss: {:.4f}, '
'envelope distance: {:.4f}, stft_l2: {:.4f}'
.format(epoch, loss_meter.average(),
envelope_meter.average(),
stft_l2_meter.average()))
history['val']['epoch'].append(epoch)
history['val']['err'].append(loss_meter.average())
history['val']['stft_l2'].append(stft_l2_meter.average())
history['val']['envelope_distance'].append(envelope_meter.average())
# Plot figure
if epoch > 0:
print('Plotting figures...')
plot_loss_metrics(args.ckpt, history)
def train(netWrapper, loader, optimizer, history, epoch, args):
torch.set_grad_enabled(True)
batch_time = AverageMeter()
data_time = AverageMeter()
# switch to train mode
netWrapper.train()
# main loop
torch.cuda.synchronize()
tic = time.perf_counter()
for i, batch_data in enumerate(loader):
# measure data time
torch.cuda.synchronize()
data_time.update(time.perf_counter() - tic)
# forward pass
netWrapper.zero_grad()
err, _ = netWrapper.forward(batch_data, args)
err = err.mean()
# backward
err.backward()
optimizer.step()
# measure total time
torch.cuda.synchronize()
batch_time.update(time.perf_counter() - tic)
tic = time.perf_counter()
# display
if i % args.disp_iter == 0:
print('Epoch: [{}][{}/{}], Time: {:.2f}, Data: {:.2f}, '
'lr_sound: {}, lr_vision: {}, '
'loss: {:.4f}'
.format(epoch, i, args.epoch_iters,
batch_time.average(), data_time.average(),
args.lr_sound, args.lr_vision,
err.item()))
fractional_epoch = epoch - 1 + 1. * i / args.epoch_iters
history['train']['epoch'].append(fractional_epoch)
history['train']['err'].append(err.item())
def checkpoint(nets, history, epoch, args):
print('Saving checkpoints at {} epochs.'.format(epoch))
(net_sound, net_vision) = nets
suffix_latest = 'latest.pth'
suffix_best = 'best.pth'
torch.save(history,
'{}/history_{}'.format(args.ckpt, suffix_latest))
torch.save(net_sound.state_dict(),
'{}/sound_{}'.format(args.ckpt, suffix_latest))
torch.save(net_vision.state_dict(),
'{}/vision_{}'.format(args.ckpt, suffix_latest))
cur_err = history['val']['err'][-1]
if cur_err < args.best_err:
args.best_err = cur_err
torch.save(net_sound.state_dict(),
'{}/sound_{}'.format(args.ckpt, suffix_best))
torch.save(net_vision.state_dict(),
'{}/vision_{}'.format(args.ckpt, suffix_best))
def create_optimizer(nets, args):
(net_sound, net_vision) = nets
param_groups = [{'params': net_sound.parameters(), 'lr': args.lr_sound},
{'params': net_vision.features.parameters(), 'lr': args.lr_vision},
{'params': net_vision.fc.parameters(), 'lr': args.lr_sound}]
return torch.optim.Adam(param_groups)
def collate_all(list_data):
# Process samples to form a batch. Point Cloud information is processed to be compatible with Minkowski Engine (ME)
# Note that ME creates a batch of point clouds coordinates putting the batch indice in the first column
# Samples come from dataset/music.py or dataset/spec.py
miscellaneous_data = []
coords = []
feats = []
for sample in list_data:
miscellaneous_data.append(sample[0])
coords_sample, points_sample, rgbs_sample, rgbs_feature = sample[-1]
coords.append(coords_sample)
if rgbs_feature:
feats.append(rgbs_sample)
else:
feats.append(points_sample)
coords_batch = ME.utils.batched_coordinates(coords)
feats_batch = torch.from_numpy(np.vstack(feats)).float()
batched_dict = torch.utils.data.dataloader.default_collate(miscellaneous_data)
batched_dict['coords'] = coords_batch
batched_dict['feats'] = feats_batch
return batched_dict
def make_data_loader(dset, args):
args = {
'batch_size': args.batch_size,
'num_workers': int(args.workers),
'collate_fn': collate_all,
'pin_memory': True,
'drop_last': False,
'shuffle': True
}
loader = torch.utils.data.DataLoader(dset, **args)
return loader
def main(args):
# Network Builders
builder = ModelBuilder()
net_sound = builder.build_sound(
arch=args.arch_sound,
visual_feature_size=args.visual_feature_size,
weights=args.weights_sound)
net_vision = builder.build_vision(
arch=args.arch_vision,
visual_feature_size=args.visual_feature_size,
weights=args.weights_vision)
nets = (net_sound, net_vision)
crit = builder.build_criterion(arch=args.loss)
# Dataset and Loader
if args.arch_sound == 'demucs':
valid_length = net_sound.valid_length(args.audLen)
delta = valid_length - args.audLen
dataset_train = MUSICWaveDataset(
args.list_train, delta, args, split='train')
dataset_val = MUSICWaveDataset(
args.list_val, delta, args, max_sample=args.num_val, split='val')
elif args.arch_sound == 'unet':
dataset_train = MUSICSpecDataset(
args.list_train, args, split='train')
dataset_val = MUSICSpecDataset(
args.list_val, args, max_sample=args.num_val, split='val')
loader_train = make_data_loader(dataset_train, args)
loader_val = make_data_loader(dataset_val, args)
args.epoch_iters = len(dataset_train) // args.batch_size
print('1 Epoch = {} iters'.format(args.epoch_iters))
# Wrap networks
netWrapper = NetWrapper(nets, crit)
netWrapper = torch.nn.DataParallel(netWrapper, device_ids=range(args.num_gpus))
netWrapper.to(args.device)
# Set up optimizer
optimizer = create_optimizer(nets, args)
# History of performance
history = {
'train': {'epoch': [], 'err': []},
'val': {'epoch': [], 'err': [], 'stft_l2': [], 'envelope_distance': []}}
# Eval mode
evaluate(netWrapper, loader_val, history, 0, args)
if args.mode == 'eval':
print('Evaluation Done!')
return
# Training loop
for epoch in range(1, args.num_epoch + 1):
train(netWrapper, loader_train, optimizer, history, epoch, args)
# Evaluation and visualization
if epoch % args.eval_epoch == 0:
evaluate(netWrapper, loader_val, history, epoch, args)
# checkpointing
checkpoint(nets, history, epoch, args)
print('Training Done!')
if __name__ == '__main__':
# arguments
parser = ArgParser()
args = parser.parse_train_arguments()
args.batch_size = args.num_gpus * args.batch_size_per_gpu
args.device = torch.device("cuda")
# experiment name
if args.mode == 'train':
args.id += '-{}mix'.format(args.num_mix)
args.id += '-{}-{}'.format(
args.arch_vision, args.arch_sound)
args.id += '-visual_feature_size{}'.format(args.visual_feature_size)
args.id += '-epoch{}'.format(args.num_epoch)
print('Model ID: {}'.format(args.id))
# paths to save/load output
args.ckpt = os.path.join(args.ckpt, args.id)
args.vis = os.path.join(args.ckpt, 'visualization/')
if args.mode == 'train':
makedirs(args.ckpt, remove=True)
elif args.mode == 'eval':
args.weights_sound = os.path.join(args.ckpt, 'sound_best.pth')
args.weights_vision = os.path.join(args.ckpt, 'vision_best.pth')
# initialize best error with a big number
args.best_err = float("inf")
random.seed(args.seed)
torch.manual_seed(args.seed)
main(args)