-
Notifications
You must be signed in to change notification settings - Fork 11
/
train.py
482 lines (398 loc) · 21.4 KB
/
train.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
"""
Author: Gurkirt Singh
Started on: 13th March 2019
Parts of this files are from many github repos
@longcw faster_rcnn_pytorch: https://github.com/longcw/faster_rcnn_pytorch
@rbgirshick py-faster-rcnn https://github.com/rbgirshick/py-faster-rcnn
Which was adopated by: Ellis Brown, Max deGroot
https://github.com/amdegroot/ssd.pytorch
mainly adopted from
https://github.com/gurkirt/realtime-action-detection
maybe more but that is where I got these from
Please don't remove above credits and give star to these repos
Licensed under The MIT License [see LICENSE for details]
"""
import os
import time
import socket
import getpass
import argparse
import datetime
import pdb
import numpy as np
import torch
import torch.nn as nn
import torch.utils.data as data_utils
from modules.solver import get_optim
from modules import utils
from modules.detection_loss import MultiBoxLoss, YOLOLoss, FocalLoss
from modules.evaluation import evaluate_detections
from modules.box_utils import decode, nms
from modules import AverageMeter
from data import DetectionDataset, custum_collate
from torchvision import transforms
from data.transforms import Resize
from models.retinanet_shared_heads import build_retinanet_shared_heads
def str2bool(v):
return v.lower() in ("yes", "true", "t", "1")
def make_01(v):
return 1 if v>0 else 0
parser = argparse.ArgumentParser(description='Training single stage FPN with OHEM, resnet as backbone')
# Name of backbone networ, e.g. resnet18, resnet34, resnet50, resnet101 resnet152 are supported
parser.add_argument('--basenet', default='resnet50', help='pretrained base model')
# if output heads are have shared features or not: 0 is no-shareing else sharining enabled
parser.add_argument('--multi_scale', default=False, type=str2bool,help='perfrom multiscale training')
parser.add_argument('--shared_heads', default=0, type=int,help='4 head layers')
parser.add_argument('--num_head_layers', default=4, type=int,help='0 mean no shareding more than 0 means shareing')
parser.add_argument('--use_bias', default=True, type=str2bool,help='0 mean no bias in head layears')
# Name of the dataset only voc or coco are supported
parser.add_argument('--dataset', default='coco', help='pretrained base model')
# Input size of image only 600 is supprted at the moment
parser.add_argument('--min_size', default=600, type=int, help='Input Size for FPN')
parser.add_argument('--max_size', default=1000, type=int, help='Input Size for FPN')
# data loading argumnets
parser.add_argument('--batch_size', default=16, type=int, help='Batch size for training')
# Number of worker to load data in parllel
parser.add_argument('--num_workers', '-j', default=4, type=int, help='Number of workers used in dataloading')
# optimiser hyperparameters
parser.add_argument('--optim', default='SGD', type=str, help='Optimiser type')
parser.add_argument('--resume', default=0, type=int, help='Resume from given iterations')
parser.add_argument('--max_iter', default=90000, type=int, help='Number of training iterations')
parser.add_argument('--lr', '--learning-rate', default=0.01, type=float, help='initial learning rate')
parser.add_argument('--momentum', default=0.9, type=float, help='momentum')
parser.add_argument('--loss_type', default='mbox', type=str, help='loss_type')
parser.add_argument('--milestones', default='60000,80000', type=str, help='Chnage the lr @')
parser.add_argument('--gammas', default='0.1,0.1', type=str, help='Gamma update for SGD')
parser.add_argument('--weight_decay', default=1e-4, type=float, help='Weight decay for SGD')
# Freeze layers or not
parser.add_argument('--fbn','--freeze_bn', default=True, type=str2bool, help='freeze bn layers if true or else keep updating bn layers')
parser.add_argument('--freezeupto', default=1, type=int, help='layer group number in ResNet up to which needs to be frozen')
# Loss function matching threshold
parser.add_argument('--positive_threshold', default=0.5, type=float, help='Min Jaccard index for matching')
parser.add_argument('--negative_threshold', default=0.4, type=float, help='Min Jaccard index for matching')
# Evaluation hyperparameters
parser.add_argument('--intial_val', default=5000, type=int, help='Initial number of training iterations before evaluation')
parser.add_argument('--val_step', default=25000, type=int, help='Number of training iterations before evaluation')
parser.add_argument('--iou_thresh', default=0.5, type=float, help='Evaluation threshold')
parser.add_argument('--conf_thresh', default=0.05, type=float, help='Confidence threshold for evaluation')
parser.add_argument('--nms_thresh', default=0.45, type=float, help='NMS threshold')
parser.add_argument('--topk', default=100, type=int, help='topk for evaluation')
# Progress logging
parser.add_argument('--log_start', default=149, type=int, help='start loging after k steps for text/Visdom/tensorboard') # Let initial ripples settle down
parser.add_argument('--log_step', default=10, type=int, help='Log every k steps for text/Visdom/tensorboard')
parser.add_argument('--tensorboard', default=False, type=str2bool, help='Use tensorboard for loss/evalaution visualization')
parser.add_argument('--visdom', default=False, type=str2bool, help='Use visdom for loss/evalaution visualization')
parser.add_argument('--vis_port', default=8098, type=int, help='Port for Visdom Server')
# Program arguments
parser.add_argument('--man_seed', default=123, type=int, help='manualseed for reproduction')
parser.add_argument('--multi_gpu', default=True, type=str2bool, help='If more than 0 then use all visible GPUs by default only one GPU used ')
# Use CUDA_VISIBLE_DEVICES=0,1,4,6 to select GPUs to use
parser.add_argument('--data_root', default='/mnt/mercury-fast/datasets/', help='Location to root directory fo dataset') # /mnt/mars-fast/datasets/
parser.add_argument('--save_root', default='/mnt/mercury-fast/datasets/', help='Location to save checkpoint models') # /mnt/sun-gamma/datasets/
parser.add_argument('--model_dir', default='', help='Location to where imagenet pretrained models exists') # /mnt/mars-fast/datasets/
## Parse arguments
args = parser.parse_args()
args = utils.set_args(args) # set directories and subsets fo datasets
if args.tensorboard:
from tensorboardX import SummaryWriter
## set random seeds and global settings
np.random.seed(args.man_seed)
torch.manual_seed(args.man_seed)
torch.cuda.manual_seed_all(args.man_seed)
torch.set_default_tensor_type('torch.FloatTensor')
def main():
args.exp_name = utils.create_exp_name(args)
args.save_root += args.dataset+'/'
args.save_root = args.save_root+'cache/'+args.exp_name+'/'
if not os.path.isdir(args.save_root): #if save directory doesn't exist create it
os.makedirs(args.save_root)
source_dir = args.save_root+'/source/' # where to save the source
utils.copy_source(source_dir)
print('\nLoading Datasets')
# ,
train_transform = transforms.Compose([
#transforms.ColorJitter(brightness=0.10, contrast=0.10, saturation=0.10, hue=0.05),
Resize(args.min_size, args.max_size),
transforms.ToTensor(),
transforms.Normalize(mean=args.means, std=args.stds)])
train_dataset = DetectionDataset(args, train=True, image_sets=args.train_sets, transform=train_transform)
print('Done Loading Dataset Train Dataset :::>>>\n',train_dataset.print_str)
val_transform = transforms.Compose([
Resize(args.min_size, args.max_size),
transforms.ToTensor(),
transforms.Normalize(mean=args.means,std=args.stds)])
val_dataset = DetectionDataset(args, train=False, image_sets=args.val_sets, transform=val_transform, full_test=False)
print('Done Loading Dataset Validation Dataset :::>>>\n',val_dataset.print_str)
args.num_classes = len(train_dataset.classes) + 1
args.classes = train_dataset.classes
args.use_bias = args.use_bias>0
args.head_size = 256
net = build_retinanet_shared_heads(args).cuda()
# print(net)
if args.multi_gpu:
print('\nLets do dataparallel\n')
net = torch.nn.DataParallel(net)
if args.fbn:
if args.multi_gpu:
net.module.backbone_net.apply(utils.set_bn_eval)
else:
net.backbone_net.apply(utils.set_bn_eval)
optimizer, scheduler, solver_print_str = get_optim(args, net)
train(args, net, optimizer, scheduler, train_dataset, val_dataset, solver_print_str)
def train(args, net, optimizer, scheduler, train_dataset, val_dataset, solver_print_str):
args.start_iteration = 0
if args.resume>100:
args.start_iteration = args.resume
args.iteration = args.start_iteration
for _ in range(args.iteration-1):
scheduler.step()
model_file_name = '{:s}/model_{:06d}.pth'.format(args.save_root, args.start_iteration)
optimizer_file_name = '{:s}/optimizer_{:06d}.pth'.format(args.save_root, args.start_iteration)
net.load_state_dict(torch.load(model_file_name))
optimizer.load_state_dict(torch.load(optimizer_file_name))
# anchors = anchors.cuda(0, non_blocking=True)
if args.tensorboard:
log_dir = args.save_root+'tensorboard-{date:%m-%d-%Hx}.log'.format(date=datetime.datetime.now())
sw = SummaryWriter(log_dir=log_dir)
log_file = open(args.save_root+'training.text{date:%m-%d-%Hx}.txt'.format(date=datetime.datetime.now()), 'w', 1)
log_file.write(args.exp_name+'\n')
for arg in sorted(vars(args)):
print(arg, getattr(args, arg))
log_file.write(str(arg)+': '+str(getattr(args, arg))+'\n')
log_file.write(str(net))
log_file.write(solver_print_str)
net.train()
# loss counters
batch_time = AverageMeter()
data_time = AverageMeter()
losses = AverageMeter()
loc_losses = AverageMeter()
cls_losses = AverageMeter()
# train_dataset = DetectionDatasetDatasetDatasetDatasetDataset(args, 'train', BaseTransform(args.input_dim, args.means, args.stds))
log_file.write(train_dataset.print_str)
log_file.write(val_dataset.print_str)
print('Train-DATA :::>>>', train_dataset.print_str)
print('VAL-DATA :::>>>', val_dataset.print_str)
epoch_size = len(train_dataset) // args.batch_size
print('Training FPN on ', train_dataset.dataset,'\n')
if args.visdom:
import visdom
viz = visdom.Visdom(env=args.exp_name, port=args.vis_port)
# initialize visdom loss plot
lot = viz.line(
X=torch.zeros((1,)).cpu(),
Y=torch.zeros((1, 6)).cpu(),
opts=dict(
xlabel='Iteration',
ylabel='Loss',
title='Training Loss',
legend=['REG', 'CLS', 'AVG', 'S-REG', ' S-CLS', ' S-AVG']
)
)
# initialize visdom meanAP and class APs plot
legends = ['meanAP']
for cls_ in args.classes:
legends.append(cls_)
print(legends)
val_lot = viz.line(
X=torch.zeros((1,)).cpu(),
Y=torch.zeros((1, args.num_classes)).cpu(),
opts=dict(
xlabel='Iteration',
ylabel='AP %',
title='Validation APs and mAP',
legend=legends
)
)
train_data_loader = data_utils.DataLoader(train_dataset, args.batch_size, num_workers=args.num_workers,
shuffle=True, pin_memory=True, collate_fn=custum_collate, drop_last=True)
val_data_loader = data_utils.DataLoader(val_dataset, args.batch_size, num_workers=args.num_workers,
shuffle=False, pin_memory=True, collate_fn=custum_collate)
torch.cuda.synchronize()
start = time.perf_counter()
iteration = args.start_iteration
eopch = 0
num_bpe = len(train_data_loader)
while iteration <= args.max_iter:
for i, (images, gts, counts, _, _) in enumerate(train_data_loader):
if iteration > args.max_iter:
break
iteration += 1
epoch = int(iteration/num_bpe)
images = images.cuda(0, non_blocking=True)
gts = gts.cuda(0, non_blocking=True)
counts = counts.cuda(0, non_blocking=True)
# forward
torch.cuda.synchronize()
data_time.update(time.perf_counter() - start)
# print(images.size(), anchors.size())
optimizer.zero_grad()
# pdb.set_trace()
# print(gts.shape, counts.shape, images.shape)
loss_l, loss_c = net(images, gts, counts)
loss_l, loss_c = loss_l.mean() , loss_c.mean()
loss = loss_l + loss_c
loss.backward()
optimizer.step()
scheduler.step()
# pdb.set_trace()
loc_loss = loss_l.item()
conf_loss = loss_c.item()
if loc_loss>300:
lline = '\n\n\n We got faulty LOCATION loss {} {} \n\n\n'.format(loc_loss, conf_loss)
log_file.write(lline)
print(lline)
loc_loss = 20.0
if conf_loss>300:
lline = '\n\n\n We got faulty CLASSIFICATION loss {} {} \n\n\n'.format(loc_loss, conf_loss)
log_file.write(lline)
print(lline)
conf_loss = 20.0
# print('Loss data type ',type(loc_loss))
loc_losses.update(loc_loss)
cls_losses.update(conf_loss)
losses.update((loc_loss + conf_loss)/2.0)
torch.cuda.synchronize()
batch_time.update(time.perf_counter() - start)
start = time.perf_counter()
if iteration % args.log_step == 0 and iteration > args.log_start:
if args.visdom:
losses_list = [loc_losses.val, cls_losses.val, losses.val, loc_losses.avg, cls_losses.avg, losses.avg]
viz.line(X=torch.ones((1, 6)).cpu() * iteration,
Y=torch.from_numpy(np.asarray(losses_list)).unsqueeze(0).cpu(),
win=lot,
update='append')
if args.tensorboard:
sw.add_scalars('Classification', {'val': cls_losses.val, 'avg':cls_losses.avg},iteration)
sw.add_scalars('Localisation', {'val': loc_losses.val, 'avg':loc_losses.avg},iteration)
sw.add_scalars('Overall', {'val': losses.val, 'avg':losses.avg},iteration)
print_line = 'Itration [{:d}]{:06d}/{:06d} loc-loss {:.2f}({:.2f}) cls-loss {:.2f}({:.2f}) ' \
'average-loss {:.2f}({:.2f}) DataTime{:0.2f}({:0.2f}) Timer {:0.2f}({:0.2f})'.format( epoch,
iteration, args.max_iter, loc_losses.val, loc_losses.avg, cls_losses.val,
cls_losses.avg, losses.val, losses.avg, 10*data_time.val, 10*data_time.avg, 10*batch_time.val, 10*batch_time.avg)
log_file.write(print_line+'\n')
print(print_line)
if iteration % (args.log_step*10) == 0:
print_line = args.exp_name
log_file.write(print_line+'\n')
print(print_line)
if (iteration % args.val_step == 0 or iteration== args.intial_val or iteration == args.max_iter) and iteration>0:
torch.cuda.synchronize()
tvs = time.perf_counter()
print('Saving state, iter:', iteration)
torch.save(net.state_dict(), '{:s}/model_{:06d}.pth'.format(args.save_root, iteration))
torch.save(optimizer.state_dict(), '{:s}/optimizer_{:06d}.pth'.format(args.save_root, iteration))
net.eval() # switch net to evaluation mode
mAP, ap_all, ap_strs, _ = validate(args, net, val_data_loader, val_dataset, iteration, iou_thresh=args.iou_thresh)
net.train()
if args.fbn:
if args.multi_gpu:
net.module.backbone_net.apply(utils.set_bn_eval)
else:
net.backbone_net.apply(utils.set_bn_eval)
for ap_str in ap_strs:
print(ap_str)
log_file.write(ap_str+'\n')
ptr_str = '\nMEANAP:::=>'+str(mAP)+'\n'
print(ptr_str)
log_file.write(ptr_str)
if args.tensorboard:
sw.add_scalar('[email protected]', mAP, iteration)
class_AP_group = dict()
for c, ap in enumerate(ap_all):
class_AP_group[args.classes[c]] = ap
sw.add_scalars('ClassAPs', class_AP_group, iteration)
if args.visdom:
aps = [mAP]
for ap in ap_all:
aps.append(ap)
viz.line(
X=torch.ones((1, args.num_classes)).cpu() * iteration,
Y=torch.from_numpy(np.asarray(aps)).unsqueeze(0).cpu(),
win=val_lot,
update='append'
)
torch.cuda.synchronize()
t0 = time.perf_counter()
prt_str = '\nValidation TIME::: {:0.3f}\n\n'.format(t0-tvs)
print(prt_str)
log_file.write(ptr_str)
log_file.close()
def validate(args, net, val_data_loader, val_dataset, iteration_num, iou_thresh=0.5):
"""Test a FPN network on an image database."""
print('Validating at ', iteration_num)
num_images = len(val_dataset)
num_classes = args.num_classes
det_boxes = [[] for _ in range(num_classes-1)]
gt_boxes = []
print_time = True
val_step = 20
count = 0
torch.cuda.synchronize()
ts = time.perf_counter()
activation = nn.Sigmoid().cuda()
if args.loss_type == 'mbox':
activation = nn.Softmax(dim=2).cuda()
with torch.no_grad():
for val_itr, (images, targets, batch_counts, img_indexs, wh) in enumerate(val_data_loader):
torch.cuda.synchronize()
t1 = time.perf_counter()
batch_size = images.size(0)
images = images.cuda(0, non_blocking=True)
decoded_boxes, conf_data = net(images)
conf_scores_all = activation(conf_data).clone()
if print_time and val_itr%val_step == 0:
torch.cuda.synchronize()
tf = time.perf_counter()
print('Forward Time {:0.3f}'.format(tf-t1))
for b in range(batch_size):
width, height = wh[b][0], wh[b][1]
gt = targets[b, :batch_counts[b]].numpy()
gt_boxes.append(gt)
# decoded_boxes = decode(loc_data[b], anchors).clone()
conf_scores = conf_scores_all[b]
#Apply nms per class and obtain the results
decoded_boxes_b = decoded_boxes[b]
for cl_ind in range(1, num_classes):
# pdb.set_trace()
scores = conf_scores[:, cl_ind].squeeze()
if args.loss_type == 'yolo':
scores = conf_scores[:, cl_ind].squeeze() * conf_scores[:, 0].squeeze() * 5.0
c_mask = scores.gt(args.conf_thresh) # greater than minmum threshold
scores = scores[c_mask].squeeze()
# print('scores size',c_mask.sum())
if scores.dim() == 0:
# print(len(''), ' dim ==0 ')
det_boxes[cl_ind - 1].append(np.asarray([]))
continue
# boxes = decoded_boxes_b.clone()
l_mask = c_mask.unsqueeze(1).expand_as(decoded_boxes_b)
boxes = decoded_boxes_b[l_mask].clone().view(-1, 4)
# idx of highest scoring and non-overlapping boxes per class
ids, counts = nms(boxes, scores, args.nms_thresh, args.topk*20) # idsn - ids after nms
scores = scores[ids[:min(args.topk,counts)]].cpu().numpy()
# pick = min(scores.shape[0], 20)
# scores = scores[:pick]
boxes = boxes[ids[:min(args.topk,counts)]].cpu().numpy()
for ik in range(boxes.shape[0]):
boxes[ik, 0] = max(0, boxes[ik, 0])
boxes[ik, 2] = min(width, boxes[ik, 2])
boxes[ik, 1] = max(0, boxes[ik, 1])
boxes[ik, 3] = min(height, boxes[ik, 3])
cls_dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=True)
det_boxes[cl_ind-1].append(cls_dets)
count += 1
if print_time and val_itr%val_step == 0:
torch.cuda.synchronize()
te = time.perf_counter()
print('im_detect: {:d}/{:d} time taken {:0.3f}'.format(count, num_images, te-ts))
torch.cuda.synchronize()
ts = time.perf_counter()
if print_time and val_itr%val_step == 0:
torch.cuda.synchronize()
te = time.perf_counter()
print('NMS stuff Time {:0.3f}'.format(te - tf))
print('Evaluating detections for itration number ', iteration_num)
return evaluate_detections(gt_boxes, det_boxes, val_dataset.classes, iou_thresh=iou_thresh)
if __name__ == '__main__':
main()