-
Notifications
You must be signed in to change notification settings - Fork 1
/
evaluate.py
585 lines (488 loc) · 22.6 KB
/
evaluate.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
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
"""
retrieve the most similar clips
"""
import os
import argparse
import pickle as pkl
import pprint
import time
import numpy as np
import torch
import cv2
import torch.nn as nn
import torch.nn.functional as F
import matplotlib.pyplot as plt
from datetime import datetime
from sklearn.metrics.pairwise import euclidean_distances, cosine_distances
from datasets import data_loader
from models.triplet_net import Tripletnet
from models.model_utils import model_selector, multipathway_input, load_checkpoint, load_pretrained_model
from datasets.data_loader import build_spatial_transformation
from datasets.temporal_transforms import TemporalCenterFrame, TemporalSpecificCrop
from datasets.temporal_transforms import Compose as TemporalCompose
import misc.distributed_helper as du_helper
from config.m_parser import load_config, arg_parser
#from misc.upload_gdrive import GoogleDriveUploader
# num_exemplar = 10
log_interval = 10
top_k = 5
exemplar_file = None
# np.random.seed(1)
# Argument parser
def m_arg_parser(parser):
parser.add_argument(
'--root_dir',
type=str,
default='.'
)
parser.add_argument(
'--name',
type=str,
default="test", #None,
help='Please specify the name (e.g. ResNet18_K, SlowFast_U): '
)
parser.add_argument(
'--num_exemplar',
type=int,
default=10,
help='Please specify number of exemplar videos: '
)
parser.add_argument(
'--heatmap',
action='store_true',
help='Run temporal heatmap visualization'
)
parser.add_argument(
"--ex_idx",
default=None,
type=int,
help='Exemplar video dataset index for the temporal heat map'
)
parser.add_argument(
"--test_idx",
default=None,
type=int,
help='Test video dataset index for the temporal heat map'
)
parser.add_argument(
"--seed",
default=0,
type=int,
help='seed for np.random'
)
parser.add_argument(
"--load_pkl",
action='store_true',
help='load computed embeddings from the pickle file'
)
parser.add_argument(
"--crop",
default='avg',
type=str,
help='avg, center, random'
)
return parser
def test_evaluate(cfg, model, cuda, device, data_loader, split='test', is_master_proc=True):
log_interval = 5
model.eval()
embedding = []
# vid_info = []
labels = []
idxs = []
world_size = du_helper.get_world_size()
def tr(x):
batch_size = x.size(0);# assert batch_size ==1
num_test_sample = x.size(2)//cfg.DATA.SAMPLE_DURATION
return x.view(cfg.DATA.INPUT_CHANNEL_NUM, num_test_sample, batch_size, cfg.DATA.SAMPLE_DURATION, cfg.DATA.SAMPLE_SIZE, cfg.DATA.SAMPLE_SIZE).permute(1,2,0,3,4,5)
with torch.no_grad():
for batch_idx, (input_seq, targets, info, indexes) in enumerate(data_loader):
if cfg.DATASET.MODALITY == True:
batch_size = input_seq[0].size(0)
if cuda:
for i in range(len(input_seq)):
input_seq[i] = input_seq[i].to(device)
else:
batch_size = input_seq.size(0)
if cuda:
input_seq= input_seq.to(device)
if cuda:
targets = targets.to(device)
indexes = indexes.to(device)
input_seq = tr(input_seq)
test_sample = input_seq.size(0)
input_seq = input_seq.squeeze(1)
embedd = model(input_seq)
if isinstance(embedd, tuple): #for multiview
embedd = embedd[0]
if cfg.NUM_GPUS > 1:
embedd, targets, indexes = du_helper.all_gather([embedd, targets, indexes])
embedd = embedd.detach().cpu()
embedding.append(embedd.mean(0))
labels.append(targets.detach().cpu())
idxs.append(indexes.detach().cpu())
batch_size_world = batch_size * world_size
if ((batch_idx + 1) * world_size) % log_interval == 0 and is_master_proc:
print('{} [{}/{} | {:.1f}%]'.format(split, (batch_idx+1)*batch_size_world, len(data_loader.dataset),
((batch_idx+1)*100.*batch_size_world/len(data_loader.dataset))))
embeddings = torch.cat(embedding, dim=0)
labels = torch.cat(labels, dim=0).tolist()
idxs = torch.cat(idxs, dim=0).tolist()
# if is_master_proc: print('embeddings size', embeddings.size())
return embeddings, labels, idxs
def evaluate(cfg, model, cuda, device, data_loader, split='train', is_master_proc=True):
#log_interval=len(data_loader.dataset)//5
log_interval = 5
model.eval()
embedding = []
# vid_info = []
labels = []
idxs = []
world_size = du_helper.get_world_size()
with torch.no_grad():
for batch_idx, (input, targets, info, indexes) in enumerate(data_loader):
if cfg.DATASET.MODALITY == True:
batch_size = input[0].size(0)
if cuda:
for i in range(len(input)):
input[i] = input[i].to(device)
else:
batch_size = input.size(0)
if cfg.MODEL.ARCH == 'slowfast':
input = multipathway_input(input, cfg)
if cuda:
for i in range(len(input)):
input[i] = input[i].to(device)
else:
if cuda:
input= input.to(device)
if cuda:
targets = targets.to(device)
indexes = indexes.to(device)
#print(input.shape)
embedd = model(input)
if cfg.MODEL.ARCH == 'coclr':
embedd = embedd[1]
else:
if isinstance(embedd, tuple): #for multiview
embedd = embedd[0]
embedd = embedd.flatten(1)
#print(embedd.shape)
if cfg.NUM_GPUS > 1:
embedd, targets, indexes = du_helper.all_gather([embedd, targets, indexes])
embedding.append(embedd.detach().cpu())
labels.append(targets.detach().cpu())
idxs.append(indexes.detach().cpu())
# vid_info.extend(info)
# print('embedd size', embedd.size())
batch_size_world = batch_size * world_size
if ((batch_idx + 1) * world_size) % log_interval == 0 and is_master_proc:
print('{} [{}/{} | {:.1f}%]'.format(split, (batch_idx+1)*batch_size_world, len(data_loader.dataset),
((batch_idx+1)*100.*batch_size_world/len(data_loader.dataset))))
embeddings = torch.cat(embedding, dim=0)
labels = torch.cat(labels, dim=0).tolist()
idxs = torch.cat(idxs, dim=0).tolist()
#if is_master_proc: print('embeddings size', embeddings.size())
return embeddings, labels, idxs
def get_distance_matrix(x_embeddings, y_embeddings=None, dist_metric='cosine'):
#print('Dist metric:', dist_metric)
assert(dist_metric in ['cosine', 'euclidean'])
if dist_metric == 'cosine':
distance_matrix = cosine_distances(x_embeddings, Y=y_embeddings)
# print(distance_matrix.size(), distance_matrix[0][0].dtype)
elif dist_metric == 'euclidean':
distance_matrix = euclidean_distances(x_embeddings, Y=y_embeddings)
# print(distance_matrix.size(), distance_matrix[0][0].dtype)
#print('Distance matrix shape:', distance_matrix.shape)
if y_embeddings is None:
np.fill_diagonal(distance_matrix, float('inf'))
return distance_matrix
def get_closest_data_mat(distance_matrix, top_k):
idx = np.argpartition(distance_matrix, top_k, axis=-1)
distance_matrix_topk_unsorted = np.take_along_axis(distance_matrix, idx[:,:top_k], axis=-1)
idx_sorted_indices = np.argsort(distance_matrix_topk_unsorted, axis=-1)
top_k = np.take_along_axis(idx, idx_sorted_indices, axis=-1)
return top_k # dim: distance_matrix.shape[0] x top_k
def get_closest_data(distance_matrix, exemplar_idx, top_k):
test_array = distance_matrix[exemplar_idx]
idx = np.argpartition(test_array, top_k)
top_k = idx[np.argsort(test_array[idx[:top_k]])]
return top_k
def plot_img(cfg, fig, val_data, train_data, num_exemplar, row, exemplar_idx, k_idx, spatial_transform=None, temporal_transform=None, output=None):
exemplar_frame = val_data._loading_img_path(exemplar_idx, temporal_transform)
test_frame = [train_data._loading_img_path(i, temporal_transform) for i in k_idx]
exemplar_title = '-'.join(exemplar_frame.split('/')[-3:-2])
print(exemplar_frame)
print('top k ids:', end=' ')
for i in k_idx:
print(i, end=' ')
pprint.pprint(test_frame)
ax = fig.add_subplot(num_exemplar,len(test_frame)+1, row*(len(test_frame)+1)+1)
image = plt.imread(exemplar_frame)
plt.imshow(image)
ax.set_title(exemplar_title, fontsize=5, pad=0.3)
plt.axis('off')
for i in range(len(test_frame)):
test_title = '-'.join(test_frame[i].split('/')[-3:-2])
ax = fig.add_subplot(num_exemplar,len(test_frame)+1, row*(len(test_frame)+1)+i+2)
image = plt.imread(test_frame[i])
plt.imshow(image)
ax.set_title(test_title, fontsize=5, pad=0.3)
plt.axis('off')
with open(os.path.join(output, 'results.txt'), 'a') as f:
f.write('exemplar_frame:\n{}\n'.format(exemplar_frame))
for frame in test_frame:
f.write(frame)
f.write('\n')
f.write('\n')
with open(os.path.join(output, 'exemplar.txt'), 'a') as f:
f.write('{}, {}'.format(exemplar_idx, exemplar_frame))
f.write('\n')
def load_exemplar(exemplar_file):
with open(exemplar_file, 'r') as f:
lines = f.readlines()
exemplar_idx = []
for line in lines:
exemplar_idx.append(int(line.split(',')[0].strip()))
return exemplar_idx
def get_topk_acc(distance_matrix, x_labels, y_labels=None, top_ks = [1,5,10,20]):
top_k = top_ks[-1]
topk_sum = 0
topk_indices = get_closest_data_mat(distance_matrix, top_k=top_k)
# print('topk_indices', topk_indices.size())
if y_labels is None:
y_labels = x_labels
acc = []
for i, x_label in enumerate(x_labels):
cur_acc = []
for k in top_ks:
topk_idx = topk_indices[:, :k]
cur_topk_idx = topk_idx[i]
topk_labels = [y_labels[j] for j in cur_topk_idx]
topk_sum = int(x_label in topk_labels)
cur_acc.append(topk_sum)
acc.append(cur_acc)
# print(acc.size())
acc = np.mean(np.array(acc), axis=0)
return acc
def get_embeddings_and_labels(args, cfg, model, cuda, device, data_loader,
split='val', is_master_proc=True, load_pkl=False, save_pkl=True):
if split == 'train':
embeddings_pkl = os.path.join(cfg.OUTPUT_PATH, 'train_embeddings.pkl')
idxs_pkl = os.path.join(cfg.OUTPUT_PATH, 'train_idxs.pkl')
labels_pkl = os.path.join(cfg.OUTPUT_PATH, 'train_labels.pkl')
else:
embeddings_pkl = os.path.join(cfg.OUTPUT_PATH, 'val_embeddings.pkl')
idxs_pkl = os.path.join(cfg.OUTPUT_PATH, 'val_idxs.pkl')
labels_pkl = os.path.join(cfg.OUTPUT_PATH, 'val_labels.pkl')
if os.path.exists(embeddings_pkl) and os.path.exists(labels_pkl) and os.path.exists(idxs_pkl) and load_pkl:
with open(embeddings_pkl, 'rb') as handle:
embeddings = torch.load(handle)
with open(labels_pkl, 'rb') as handle:
labels = torch.load(handle)
with open(idxs_pkl, 'rb') as handle:
idxs = torch.load(handle)
print('retrieved {}_embeddings'.format(split), embeddings.size(), 'labels', len(labels))
else:
if split =="test": #COCLR evaluation
embeddings, labels, idxs = test_evaluate(cfg, model, cuda, device, data_loader, split=split, is_master_proc=is_master_proc)
else:
embeddings, labels, idxs = evaluate(cfg, model, cuda, device, data_loader, split=split, is_master_proc=is_master_proc)
if save_pkl and is_master_proc:
with open(embeddings_pkl, 'wb') as handle:
torch.save(embeddings, handle, pickle_protocol=pkl.HIGHEST_PROTOCOL)
with open(labels_pkl, 'wb') as handle:
torch.save(labels, handle, pickle_protocol=pkl.HIGHEST_PROTOCOL)
with open(idxs_pkl, 'wb') as handle:
torch.save(idxs, handle, pickle_protocol=pkl.HIGHEST_PROTOCOL)
print('saved {}_embeddings'.format(split), embeddings.size(), 'labels', len(labels))
if split=="test":
embeddings = embeddings.reshape((-1, cfg.LOSS.FEAT_DIM))
return embeddings, labels, idxs
def k_nearest_embeddings(args, model, cuda, device, train_loader, test_loader,
train_data, val_data, cfg, test_split='val', plot=True, epoch=None, is_master_proc=True,
evaluate_output=None, num_exemplar=None, service=None,
load_pkl=False, out_filename='global_retrieval_acc'):
if is_master_proc:
print ('Getting embeddings...')
test_embeddings, test_labels, _ = get_embeddings_and_labels(args, cfg, model, cuda, device, test_loader,
split=test_split, is_master_proc=is_master_proc, load_pkl=load_pkl)
train_embeddings, train_labels, _ = get_embeddings_and_labels(args, cfg, model, cuda, device, train_loader,
split='train', is_master_proc=is_master_proc, load_pkl=load_pkl)
acc = []
print ('Computing top1/5/10/20 Acc...')
if (is_master_proc):
distance_matrix = get_distance_matrix(test_embeddings, train_embeddings, dist_metric=cfg.LOSS.DIST_METRIC)
acc = get_topk_acc(distance_matrix, test_labels, y_labels=train_labels)
if epoch is not None:
to_write = 'epoch:{} {:.2f} {:.2f}'.format(epoch, 100.*acc[0], 100.*acc[1], 100.*acc[2], 100.*acc[3])
msg = '\nTest Set: Top1 Acc: {:.2f}%, Top5 Acc: {:.2f}%, Top10 Acc: {:.2f}%, Top20 Acc: {:.2f}%'.format(100.*acc[0], 100.*acc[1], 100.*acc[2], 100.*acc[3])
to_write += '\n'
with open('{}/tnet_checkpoints/{}.txt'.format(cfg.OUTPUT_PATH, out_filename), "a") as val_file:
val_file.write(to_write)
if plot:
spatial_transform = build_spatial_transformation(cfg, 'val', triplets=False)
temporal_transform = [TemporalCenterFrame()]
temporal_transform = TemporalCompose(temporal_transform)
fig = plt.figure()
for i in range(num_exemplar):
exemplar_idx = np.random.randint(0, distance_matrix.shape[0]-1)
print('exemplar video id: {}, label:{}'.format(exemplar_idx, test_labels[exemplar_idx]))
k_idx = get_closest_data(distance_matrix, exemplar_idx, top_k)
print([train_labels[x] for x in k_idx], len(train_data))
k_nearest_data = [train_data[i] for i in k_idx]
plot_img(cfg, fig, val_data, train_data, num_exemplar, i, exemplar_idx, k_idx, spatial_transform, temporal_transform, output=evaluate_output)
# plt.show()
png_file = os.path.join(evaluate_output, '{}_plot.png'.format(os.path.basename(evaluate_output)))
fig.tight_layout(pad=3.5)
plt.savefig(png_file, dpi=300)
if service is not None:
service.upload_file_to_gdrive(png_file, 'evaluate')
print('figure saved to: {}, and uploaded to GoogleDrive'.format(png_file))
# print(acc)
print('Top1 Acc: {:.2f}%, Top5 Acc: {:.2f}%, Top10 Acc: {:.2f}%, Top20 Acc: {:.2f}%'.format(100.*acc[0], 100.*acc[1], 100.*acc[2], 100.*acc[3]))
return acc
def temporal_heat_map(model, data, cfg, evaluate_output, exemplar_idx=455,
test_idx=456):
num_frames_exemplar = data.data[exemplar_idx]['num_frames']
exemplar_video_full, _, _ = data._get_video_custom_temporal(exemplar_idx) # full size
exemplar_video_full = exemplar_video_full.unsqueeze(0)
num_frames_crop = cfg.DATA.SAMPLE_DURATION
stride = num_frames_crop // 2
dists = []
model.eval()
with torch.no_grad():
test_video, _, _ = data.__getitem__(test_idx) # cropped size
test_video = test_video.unsqueeze(0)
print('Test input size:', test_video.size(), '\n')
if (cfg.MODEL.ARCH == 'slowfast'):
test_video_in = multipathway_input(test_video, cfg)
if cuda:
for i in range(len(test_video_in)):
test_video_in[i] = test_video_in[i].to(device)
else:
if cuda:
test_video_in = test_video_in.to(device)
test_embedding = model(test_video_in)
if isinstance(test_embedding, tuple):
test_embedding = test_embedding[0]
#print('Test embed size:', test_embedding.size())
# Loop across exemplar video, use [i-cfg.DATA.SAMPLE_SIZE,...,i] as the frames for the temporal crop
for i in range(num_frames_crop, num_frames_exemplar, stride):
temporal_transform_exemplar = TemporalSpecificCrop(begin_index=i-num_frames_crop, size=num_frames_crop)
exemplar_video, _, _ = data._get_video_custom_temporal(exemplar_idx, temporal_transform_exemplar) # full siz
exemplar_video = exemplar_video.unsqueeze(0)
if (cfg.MODEL.ARCH == 'slowfast'):
exemplar_video_in = multipathway_input(exemplar_video, cfg)
if cuda:
for j in range(len(exemplar_video_in)):
exemplar_video_in[j] = exemplar_video_in[j].to(device)
else:
if cuda:
exemplar_video_in = exemplar_video_in.to(device)
exemplar_embedding = model(exemplar_video_in)
if isinstance(exemplar_embedding, tuple):
exemplar_embedding = exemplar_embedding[0]
dist = F.pairwise_distance(exemplar_embedding, test_embedding, 2)
dists.append(dist.item())
#print(dists)
x = []
y = []
plt.show()
axes = plt.gca()
axes.set_xlim(0, num_frames_exemplar)
axes.set_ylim(0, max(dists))
line, = axes.plot(x, y, 'b-')
dist_idx = 0
# channels x frames x width, height --> frames x width x height x channels
video_ex = exemplar_video_full[0].permute(1,2,3,0)
video_test = test_video[0].permute(1,2,3,0)
fps = 25.0
for i in range(len(video_ex)):
blank_divider = np.full((2,128,3), 256, dtype=int)
np_vertical_stack = np.vstack(( video_ex[i].numpy(), blank_divider, video_test[i % len(video_test)].numpy() ))
cv2.imshow('Videos', np_vertical_stack)
# show plot of embedding distance for past num_frames_crop frames of exemplar video
if i >= num_frames_crop and (i - num_frames_crop) % stride == 0:
x.append(i)
y.append(dists[dist_idx])
line.set_xdata(x)
line.set_ydata(y)
plt.draw()
plt.pause(1e-17)
dist_idx += 1
cv2.waitKey(int(1.0/fps*1000.0))
if __name__ == '__main__':
args = m_arg_parser(arg_parser()).parse_args()
cfg = load_config(args)
print("Train batch size:", cfg.TRAIN.BATCH_SIZE)
print("Val batch size:", cfg.VAL.BATCH_SIZE)
np.random.seed(args.seed)
force_data_parallel = True
os.environ["CUDA_VISIBLE_DEVICES"]=str(args.gpu)
global cuda; cuda = torch.cuda.is_available()
global device; device = torch.device("cuda" if cuda else "cpu")
name = args.name
num_exemplar = args.num_exemplar
if not name:
name = input('Please specify the name (e.g. ResNet18_K, SlowFast_U): ')
if not num_exemplar:
num_exemplar = int(input('Please specify number of exemplar videos: '))
if not cfg.OUTPUT_PATH:
output = input('Please specify output directory: ')
cfg.OUTPUT_PATH = output
else:
output = cfg.OUTPUT_PATH
start = time.time()
now = datetime.now()
evaluate_output = os.path.join(output, '{}_evaluate'.format(name))
if not os.path.exists(evaluate_output):
os.makedirs(evaluate_output)
print('made output dir:{}'.format(evaluate_output))
# ============================== Model Setup ===============================
# Check if this is the master process (true if not distributed)
is_master_proc = du_helper.is_master_proc(cfg.NUM_GPUS)
# Select appropriate model
if(is_master_proc):
print('\n==> Generating {} backbone model...'.format(cfg.MODEL.ARCH))
model=model_selector(cfg, projection_head=True)
## SyncBatchNorm
if cfg.SYNC_BATCH_NORM:
print('Converting BatchNorm*D to SyncBatchNorm!')
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
n_parameters = sum([p.data.nelement() for p in model.parameters()])
if(is_master_proc):
print('Number of params: {}'.format(n_parameters))
# Load similarity network checkpoint if path exists
if args.checkpoint_path is not None:
start_epoch, best_acc = load_checkpoint(model, args.checkpoint_path, is_master_proc)
if cuda:
#model = DDP(model)
if torch.cuda.device_count() > 1:
print("Using DataParallel with {} gpus".format(torch.cuda.device_count()))
model = nn.DataParallel(model)
model = model.cuda(device=device)
if args.pretrain_path is not None:
model = load_pretrained_model(model, args.pretrain_path, is_master_proc)
print('=> finished generating similarity network...')
cfg.DATA.TEMPORAL_CROP = args.crop
print('=> Using evaluation method: {}'.format(cfg.DATA.TEMPORAL_CROP))
if args.crop=='avg': #COCLR eval metric
test_split='test'
else:
test_split='val'
# ============================== Data Loaders ==============================
train_loader, (train_data, _) = data_loader.build_data_loader('train', cfg, triplets=False, req_train_shuffle=False)
test_loader, (val_data, _) = data_loader.build_data_loader(test_split, cfg, triplets=False, val_sample=None, req_train_shuffle=False)
# ================================ Evaluate ================================
if args.heatmap:
if args.ex_idx and args.test_idx:
temporal_heat_map(model, data, cfg, evaluate_output, args.ex_idx,
args.test_idx)
else:
print ('No exemplar and test indices provided')
temporal_heat_map(model, data, cfg, evaluate_output)
else:
k_nearest_embeddings(args, model, cuda, device, train_loader, test_loader,
train_data, val_data, cfg, test_split=test_split, evaluate_output=evaluate_output,
num_exemplar=num_exemplar, load_pkl=args.load_pkl)
print('total runtime: {}s'.format(time.time()-start))