def main():
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
use_gpu = torch.cuda.is_available()
sys.stdout = Logger(osp.join(args.resume, 'log_test.txt'))
print("==========\nArgs:{}\n==========".format(args))
print("Initializing dataset {}".format(args.dataset))
dataset = data_manager.init_dataset(name=args.dataset, root=args.root)
# Data augmentation
spatial_transform_test = ST.Compose([
ST.Scale((args.height, args.width), interpolation=3),
ST.ToTensor(),
ST.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
temporal_transform_test = None
pin_memory = True if use_gpu else False
queryloader = DataLoader(VideoDataset(
dataset.query,
spatial_transform=spatial_transform_test,
temporal_transform=temporal_transform_test),
batch_size=1,
shuffle=False,
num_workers=0,
pin_memory=pin_memory,
drop_last=False)
galleryloader = DataLoader(VideoDataset(
dataset.gallery,
spatial_transform=spatial_transform_test,
temporal_transform=temporal_transform_test),
batch_size=1,
shuffle=False,
num_workers=0,
pin_memory=pin_memory,
drop_last=False)
print("Initializing model: {}".format(args.arch))
model = models.init_model(name=args.arch,
num_classes=dataset.num_train_pids)
print("Model size: {:.5f}M".format(
sum(p.numel() for p in model.parameters()) / 1000000.0))
for epoch in args.test_epochs:
model_path = osp.join(args.resume,
'checkpoint_ep' + str(epoch) + '.pth.tar')
print("Loading checkpoint from '{}'".format(model_path))
checkpoint = torch.load(model_path)
model.load_state_dict(checkpoint['state_dict'])
if use_gpu: model = model.cuda()
print("Evaluate")
with torch.no_grad():
test(model, queryloader, galleryloader, use_gpu)
def getDataSets(dataset):
spatial_transform_test = ST.Compose([
ST.Scale((args.height, args.width), interpolation=3),
ST.ToNumpy(),
ST.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
temporal_transform_test = TT.TemporalBeginCrop()
queryset = VideoDataset(dataset.query, spatial_transform=spatial_transform_test, temporal_transform=temporal_transform_test)
galleryset =VideoDataset(dataset.gallery, spatial_transform=spatial_transform_test, temporal_transform=temporal_transform_test)
return queryset,galleryset
def getDataSets(dataset):
spatial_transform_train = ST.Compose([
ST.Scale((args.height, args.width), interpolation=3),
ST.RandomHorizontalFlip(),
ST.ToNumpy(),
ST.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
temporal_transform_train = TT.TemporalRandomCrop(size=args.seq_len, stride=args.sample_stride)
spatial_transform_test = ST.Compose([
ST.Scale((args.height, args.width), interpolation=3),
ST.ToNumpy(),
ST.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
temporal_transform_test = TT.TemporalBeginCrop()
if args.dataset != 'mars':
trainset = VideoDataset(dataset.train_dense, spatial_transform=spatial_transform_train, temporal_transform=temporal_transform_train)
else:
trainset = VideoDataset(dataset.train, spatial_transform=spatial_transform_train, temporal_transform=temporal_transform_train)
queryset = VideoDataset(dataset.query, spatial_transform=spatial_transform_test, temporal_transform=temporal_transform_test)
galleryset =VideoDataset(dataset.gallery, spatial_transform=spatial_transform_test, temporal_transform=temporal_transform_test)
return trainset,queryset,galleryset
def train(parameters: Dict[str, float]) -> nn.Module:
global args
print("====", args.focus, "=====")
torch.manual_seed(args.seed)
# args.gpu_devices = "0,1"
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_devices
use_gpu = torch.cuda.is_available()
cudnn.benchmark = True
torch.cuda.manual_seed_all(args.seed)
dataset = data_manager.init_dataset(name=args.dataset, sampling= args.sampling)
transform_test = transforms.Compose([
transforms.Resize((args.height, args.width), interpolation=3),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
pin_memory = True if use_gpu else False
transform_train = transforms.Compose([
transforms.Resize((args.height, args.width), interpolation=3),
transforms.RandomHorizontalFlip(p=0.5),
transforms.Pad(10),
Random2DTranslation(args.height, args.width),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
batch_size = int(round(parameters.get("batch_size", 32) ))
base_learning_rate = 0.00035
# weight_decay = 0.0005
alpha = parameters.get("alpha", 1.2)
sigma = parameters.get("sigma", 0.8)
l = parameters.get("l", 0.5)
beta_ratio = parameters.get("beta_ratio", 0.5)
gamma = parameters.get("gamma", 0.1)
margin = parameters.get("margin", 0.3)
weight_decay = parameters.get("weight_decay", 0.0005)
lamb = 0.3
num_instances = 4
pin_memory = True
trainloader = DataLoader(
VideoDataset(dataset.train, seq_len=args.seq_len, sample='random',transform=transform_train),
sampler=RandomIdentitySampler(dataset.train, num_instances=args.num_instances),
batch_size=batch_size, num_workers=args.workers,
pin_memory=pin_memory, drop_last=True,
)
if args.dataset == 'mars_subset' :
validation_loader = DataLoader(
VideoDataset(dataset.val, seq_len=8, sample='random', transform=transform_test),
batch_size=args.test_batch, shuffle=False, num_workers=args.workers,
pin_memory=pin_memory, drop_last=False,
)
else:
queryloader = DataLoader(
VideoDataset(dataset.val_query, seq_len=args.seq_len, sample='dense_subset', transform=transform_test),
batch_size=args.test_batch, shuffle=False, num_workers=args.workers,
pin_memory=pin_memory, drop_last=False,
)
galleryloader = DataLoader(
VideoDataset(dataset.val_gallery, seq_len=args.seq_len, sample='dense_subset', transform=transform_test),
batch_size=args.test_batch, shuffle=False, num_workers=args.workers,
pin_memory=pin_memory, drop_last=False,
)
criterion_htri = TripletLoss(margin, 'cosine')
criterion_xent = CrossEntropyLabelSmooth(dataset.num_train_pids)
criterion_center_loss = CenterLoss(use_gpu=1)
criterion_osm_caa = OSM_CAA_Loss(alpha=alpha , l=l , osm_sigma=sigma )
args.arch = "ResNet50ta_bt"
model = models.init_model(name=args.arch, num_classes=dataset.num_train_pids, loss={'xent', 'htri'})
if use_gpu:
model = nn.DataParallel(model).cuda()
params = []
for key, value in model.named_parameters():
if not value.requires_grad:
continue
lr = base_learning_rate
weight_decay = weight_decay
params += [{"params": [value], "lr": lr, "weight_decay": weight_decay}]
optimizer = torch.optim.Adam(params)
scheduler = WarmupMultiStepLR(optimizer, milestones=[40, 70], gamma=gamma, warmup_factor=0.01, warmup_iters=10)
optimizer_center = torch.optim.SGD(criterion_center_loss.parameters(), lr=0.5)
start_epoch = args.start_epoch
best_rank1 = -np.inf
num_epochs = 121
if 'mars' not in args.dataset :
num_epochs = 121
# test_rerank(model, queryloader, galleryloader, args.pool, use_gpu, lamb=lamb , parameters=parameters)
for epoch in range (num_epochs):
vals = train_model(model, criterion_xent, criterion_htri, optimizer, trainloader, use_gpu , optimizer_center , criterion_center_loss, criterion_osm_caa, beta_ratio)
if math.isnan(vals[0]):
return 0
scheduler.step()
if epoch % 40 ==0 :
print("TripletLoss {:.6f} OSM Loss {:.6f} Cross_entropy {:.6f} Total Loss {:.6f} ".format(vals[1] , vals[3] , vals[1] , vals[0]))
if args.dataset == 'mars_subset' :
result1 = test_validation(model, validation_loader, args.pool, use_gpu, parameters=parameters)
del validation_loader
else:
result1= test_rerank(model, queryloader, galleryloader, args.pool, use_gpu, lamb=lamb , parameters=parameters)
del queryloader
del galleryloader
del trainloader
del model
del criterion_htri
del criterion_xent
del criterion_center_loss
del criterion_osm_caa
del optimizer
del optimizer_center
del scheduler
return result1
Random2DTranslation(args.height, args.width),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224,
0.225]),
RandomErasing2(probability=0.5, mean=[0.485, 0.456, 0.406])
])
transform_test = transforms.Compose([
transforms.Resize((args.height, args.width), interpolation=3),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
trainloader = DataLoader(
VideoDataset(dataset.train,
seq_len=args.seq_len,
sample='random',
transform=transform_train),
sampler=RandomIdentitySampler(dataset.train,
num_instances=args.num_instances),
batch_size=args.train_batch,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=True,
)
queryloader = DataLoader(
VideoDataset(dataset.query,
seq_len=args.seq_len,
sample='dense',
transform=transform_test),
batch_size=args.test_batch,
def main():
torch.manual_seed(args.seed)
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
use_gpu = torch.cuda.is_available()
if args.use_cpu: use_gpu = False
sys.stdout = Logger(osp.join(args.save_dir, 'log_train.txt'))
print("==========\nArgs:{}\n==========".format(args))
if use_gpu:
print("Currently using GPU {}".format(args.gpu))
torch.cuda.manual_seed_all(args.seed)
else:
print("Currently using CPU (GPU is highly recommended)")
print("Initializing dataset {}".format(args.dataset))
dataset = data_manager.init_dataset(name=args.dataset, root=args.root)
# Data augmentation
spatial_transform_train = ST.Compose([
ST.Scale((args.height, args.width), interpolation=3),
ST.RandomHorizontalFlip(),
ST.ToTensor(),
ST.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
temporal_transform_train = TT.TemporalRandomCrop(size=args.seq_len,
stride=args.sample_stride)
spatial_transform_test = ST.Compose([
ST.Scale((args.height, args.width), interpolation=3),
ST.ToTensor(),
ST.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
temporal_transform_test = TT.TemporalBeginCrop()
pin_memory = True if use_gpu else False
if args.dataset != 'mars':
trainloader = DataLoader(
VideoDataset(dataset.train_dense,
spatial_transform=spatial_transform_train,
temporal_transform=temporal_transform_train),
sampler=RandomIdentitySampler(dataset.train_dense,
num_instances=args.num_instances),
batch_size=args.train_batch,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=True)
else:
trainloader = DataLoader(
VideoDataset(dataset.train,
spatial_transform=spatial_transform_train,
temporal_transform=temporal_transform_train),
sampler=RandomIdentitySampler(dataset.train,
num_instances=args.num_instances),
batch_size=args.train_batch,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=True)
queryloader = DataLoader(VideoDataset(
dataset.query,
spatial_transform=spatial_transform_test,
temporal_transform=temporal_transform_test),
batch_size=args.test_batch,
shuffle=False,
num_workers=0,
pin_memory=pin_memory,
drop_last=False)
galleryloader = DataLoader(VideoDataset(
dataset.gallery,
spatial_transform=spatial_transform_test,
temporal_transform=temporal_transform_test),
batch_size=args.test_batch,
shuffle=False,
num_workers=0,
pin_memory=pin_memory,
drop_last=False)
print("Initializing model: {}".format(args.arch))
model = models.init_model(name=args.arch,
num_classes=dataset.num_train_pids)
print("Model size: {:.5f}M".format(
sum(p.numel() for p in model.parameters()) / 1000000.0))
criterion_xent = nn.CrossEntropyLoss()
criterion_htri = TripletLoss(margin=args.margin, distance=args.distance)
optimizer = torch.optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
scheduler = lr_scheduler.MultiStepLR(optimizer,
milestones=args.stepsize,
gamma=args.gamma)
start_epoch = args.start_epoch
if args.resume:
print("Loading checkpoint from '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
start_epoch = checkpoint['epoch']
if use_gpu:
model = nn.DataParallel(model).cuda()
# model = model.cuda()
start_time = time.time()
train_time = 0
best_rank1 = -np.inf
best_epoch = 0
print("==> Start training")
for epoch in range(start_epoch, args.max_epoch):
scheduler.step()
start_train_time = time.time()
train(epoch, model, criterion_xent, criterion_htri, optimizer,
trainloader, use_gpu)
train_time += round(time.time() - start_train_time)
if (epoch + 1) >= args.start_eval and args.eval_step > 0 and (
epoch + 1) % args.eval_step == 0 or (epoch +
1) == args.max_epoch:
print("==> Test")
with torch.no_grad():
# test using 4 frames
rank1 = test(model, queryloader, galleryloader, use_gpu)
is_best = rank1 > best_rank1
if is_best:
best_rank1 = rank1
best_epoch = epoch + 1
if use_gpu:
state_dict = model.module.state_dict()
else:
state_dict = model.state_dict()
save_checkpoint(
{
'state_dict': state_dict,
'rank1': rank1,
'epoch': epoch,
}, is_best,
osp.join(args.save_dir,
'checkpoint_ep' + str(epoch + 1) + '.pth.tar'))
print("==> Best Rank-1 {:.1%}, achieved at epoch {}".format(
best_rank1, best_epoch))
elapsed = round(time.time() - start_time)
elapsed = str(datetime.timedelta(seconds=elapsed))
train_time = str(datetime.timedelta(seconds=train_time))
print(
"Finished. Total elapsed time (h:m:s): {}. Training time (h:m:s): {}.".
format(elapsed, train_time))