def get_data(name,
data_dir,
height,
width,
batch_size,
workers,
combine_trainval,
crop,
tracking_icams,
fps,
real=True,
synthetic=True,
re=0,
num_instances=0,
camstyle=0,
zju=0,
colorjitter=0):
root = osp.join(data_dir, name)
if name == 'duke_tracking':
if tracking_icams != 0:
tracking_icams = [tracking_icams]
else:
tracking_icams = list(range(1, 9))
dataset = datasets.create(name,
root,
type='tracking_gt',
iCams=tracking_icams,
fps=fps,
trainval=combine_trainval)
elif name == 'aic_tracking':
dataset = datasets.create(name,
root,
type='tracking_gt',
fps=fps,
trainval=combine_trainval)
else:
dataset = datasets.create(name, root, real, synthetic)
normalizer = T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
num_classes = dataset.num_train_ids
train_transformer = T.Compose([
T.ColorJitter(brightness=0.1 * colorjitter,
contrast=0.1 * colorjitter,
saturation=0.1 * colorjitter,
hue=0),
T.Resize((height, width)),
T.RandomHorizontalFlip(),
T.Pad(10 * crop),
T.RandomCrop((height, width)),
T.ToTensor(),
normalizer,
T.RandomErasing(probability=re),
])
test_transformer = T.Compose([
T.Resize((height, width)),
# T.RectScale(height, width, interpolation=3),
T.ToTensor(),
normalizer,
])
if zju:
train_loader = DataLoader(Preprocessor(dataset.train,
root=dataset.train_path,
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
sampler=ZJU_RandomIdentitySampler(
dataset.train, batch_size, num_instances)
if num_instances else None,
shuffle=False if num_instances else True,
pin_memory=True,
drop_last=False if num_instances else True)
else:
train_loader = DataLoader(
Preprocessor(dataset.train,
root=dataset.train_path,
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
sampler=RandomIdentitySampler(dataset.train, num_instances)
if num_instances else None,
shuffle=False if num_instances else True,
pin_memory=True,
drop_last=True)
query_loader = DataLoader(Preprocessor(dataset.query,
root=dataset.query_path,
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
gallery_loader = DataLoader(Preprocessor(dataset.gallery,
root=dataset.gallery_path,
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
if camstyle <= 0:
camstyle_loader = None
else:
camstyle_loader = DataLoader(Preprocessor(dataset.camstyle,
root=dataset.camstyle_path,
transform=train_transformer),
batch_size=camstyle,
num_workers=workers,
shuffle=True,
pin_memory=True,
drop_last=True)
return dataset, num_classes, train_loader, query_loader, gallery_loader, camstyle_loader
def main(args):
np.random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.benchmark = True
# Create data loaders
assert args.num_instances > 1, "num_instances should be greater than 1"
assert args.batch_size % args.num_instances == 0, \
'num_instances should divide batch_size'
if args.height is None or args.width is None:
args.height, args.width = (144, 56) if args.arch == 'inception' else \
(256, 128)
# get_source_data
src_dataset, src_extfeat_loader = \
get_source_data(args.src_dataset, args.data_dir, args.height,
args.width, args.batch_size, args.workers)
# get_target_data
tgt_dataset, num_classes, tgt_extfeat_loader, test_loader = \
get_data(args.tgt_dataset, args.data_dir, args.height,
args.width, args.batch_size, args.workers)
# Create model
# Hacking here to let the classifier be the number of source ids
if args.src_dataset == 'dukemtmc':
model = models.create(args.arch, num_classes=632, pretrained=False)
elif args.src_dataset == 'market1501':
model = models.create(args.arch, num_classes=676, pretrained=False)
else:
raise RuntimeError(
'Please specify the number of classes (ids) of the network.')
# Load from checkpoint
start_epoch = best_top1 = 0
if args.resume:
print(
'Resuming checkpoints from finetuned model on another dataset...\n'
)
checkpoint = load_checkpoint(args.resume)
model.load_state_dict(checkpoint['state_dict'], strict=False)
else:
raise RuntimeWarning('Not using a pre-trained model.')
model = nn.DataParallel(model).cuda()
# Distance metric
metric = DistanceMetric(algorithm=args.dist_metric)
# Evaluator
evaluator = Evaluator(model, print_freq=args.print_freq)
print("Test with the original model trained on source domain:")
evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
if args.evaluate:
return
# Criterion
criterion = []
criterion.append(
TripletLoss(margin=args.margin,
num_instances=args.num_instances).cuda())
criterion.append(
TripletLoss(margin=args.margin,
num_instances=args.num_instances).cuda())
# Optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=args.lr, momentum=0.9)
# training stage transformer on input images
normalizer = T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_transformer = T.Compose([
T.Resize((args.height, args.width)),
T.RandomHorizontalFlip(),
T.ToTensor(), normalizer,
T.RandomErasing(probability=0.5, sh=0.2, r1=0.3)
])
# Start training
for iter_n in range(args.iteration):
if args.lambda_value == 0:
source_features = 0
else:
# get source datas' feature
source_features, _ = extract_features(model,
src_extfeat_loader,
print_freq=args.print_freq)
# synchronization feature order with src_dataset.train
source_features = torch.cat([
source_features[f].unsqueeze(0)
for f, _, _ in src_dataset.train
], 0)
# extract training images' features
print('Iteration {}: Extracting Target Dataset Features...'.format(
iter_n + 1))
target_features, _ = extract_features(model,
tgt_extfeat_loader,
print_freq=args.print_freq)
# synchronization feature order with dataset.train
target_features = torch.cat(
[target_features[f].unsqueeze(0) for f, _, _ in tgt_dataset.train],
0)
# calculate distance and rerank result
print('Calculating feature distances...')
target_features = target_features.numpy()
rerank_dist = re_ranking(source_features,
target_features,
lambda_value=args.lambda_value)
if iter_n == 0:
# DBSCAN cluster
tri_mat = np.triu(rerank_dist, 1) # tri_mat.dim=2
tri_mat = tri_mat[np.nonzero(tri_mat)] # tri_mat.dim=1
tri_mat = np.sort(tri_mat, axis=None)
top_num = np.round(args.rho * tri_mat.size).astype(int)
eps = tri_mat[:top_num].mean()
print('eps in cluster: {:.3f}'.format(eps))
cluster = DBSCAN(eps=eps,
min_samples=4,
metric='precomputed',
n_jobs=8)
# select & cluster images as training set of this epochs
print('Clustering and labeling...')
labels = cluster.fit_predict(rerank_dist)
num_ids = len(set(labels)) - 1
print('Iteration {} have {} training ids'.format(iter_n + 1, num_ids))
# generate new dataset
new_dataset = []
for (fname, _, _), label in zip(tgt_dataset.trainval, labels):
if label == -1:
continue
# dont need to change codes in trainer.py _parsing_input function and sampler function after add 0
new_dataset.append((fname, label, 0))
print('Iteration {} have {} training images'.format(
iter_n + 1, len(new_dataset)))
train_loader = DataLoader(Preprocessor(new_dataset,
root=tgt_dataset.images_dir,
transform=train_transformer),
batch_size=args.batch_size,
num_workers=4,
sampler=RandomIdentitySampler(
new_dataset, args.num_instances),
pin_memory=True,
drop_last=True)
# train model with new generated dataset
trainer = Trainer(model, criterion, print_freq=args.print_freq)
evaluator = Evaluator(model, print_freq=args.print_freq)
# Start training
for epoch in range(args.epochs):
trainer.train(epoch, train_loader, optimizer)
# Evaluate
rank1 = evaluator.evaluate(test_loader, tgt_dataset.query,
tgt_dataset.gallery)
save_checkpoint(
{
'state_dict': model.module.state_dict(),
'epoch': iter_n + 1,
'num_ids': num_ids,
},
True,
fpath=osp.join(args.logs_dir, 'checkpoint.pth.tar'))
print('\n * Finished epoch {:3d} rank1: {:5.1%} \n'.format(
iter_n + 1, rank1))
def main(args):
np.random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.benchmark = True
# Create data loaders
assert args.num_instances > 1, "num_instances should be greater than 1"
assert args.batch_size % args.num_instances == 0, \
'num_instances should divide batch_size'
if args.height is None or args.width is None:
args.height, args.width = (144, 56) if args.arch == 'inception' else \
(256, 128)
# get source data
src_dataset, src_extfeat_loader = \
get_source_data(args.src_dataset, args.data_dir, args.height,
args.width, args.batch_size, args.workers)
# get target data
tgt_dataset, num_classes, tgt_extfeat_loader, test_loader = \
get_data(args.tgt_dataset, args.data_dir, args.height,
args.width, args.batch_size, args.workers)
# Create model
# Hacking here to let the classifier be the number of source ids
if args.src_dataset == 'dukemtmc':
model = models.create(args.arch, num_classes=632, pretrained=False)
elif args.src_dataset == 'market1501':
model = models.create(args.arch, num_classes=676, pretrained=False)
else:
raise RuntimeError('Please specify the number of classes (ids) of the network.')
# Load from checkpoint
if args.resume:
print('Resuming checkpoints from finetuned model on another dataset...\n')
checkpoint = load_checkpoint(args.resume)
model.load_state_dict(checkpoint['state_dict'], strict=False)
else:
raise RuntimeWarning('Not using a pre-trained model.')
model = nn.DataParallel(model).cuda()
# evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
# if args.evaluate: return
# Criterion
criterion = [
TripletLoss(args.margin, args.num_instances, isAvg=True, use_semi=True).cuda(),
TripletLoss(args.margin, args.num_instances, isAvg=True, use_semi=True).cuda(),
None
]
# Optimizer
optimizer = torch.optim.Adam(
model.parameters(), lr=args.lr
)
# training stage transformer on input images
normalizer = T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
train_transformer = T.Compose([
T.Resize((args.height,args.width)),
T.RandomHorizontalFlip(),
T.ToTensor(), normalizer,
T.RandomErasing(probability=0.5, sh=0.2, r1=0.3)
])
# # Start training
for iter_n in range(args.iteration):
if args.lambda_value == 0:
source_features = 0
else:
# get source datas' feature
source_features, _ = extract_features(model, src_extfeat_loader, print_freq=args.print_freq)
# synchronization feature order with src_dataset.train
source_features = torch.cat([source_features[f].unsqueeze(0) for f, _, _, _ in src_dataset.train], 0)
# extract training images' features
print('Iteration {}: Extracting Target Dataset Features...'.format(iter_n+1))
target_features, tarNames = extract_features(model, tgt_extfeat_loader, print_freq=args.print_freq)
# synchronization feature order with dataset.train
target_features = torch.cat([target_features[f].unsqueeze(0) for f, _, _, _ in tgt_dataset.trainval], 0)
target_real_label = np.asarray([tarNames[f].unsqueeze(0) for f, _, _, _ in tgt_dataset.trainval])
numTarID = len(set(target_real_label))
print('n_clusters =', numTarID)
# calculate distance and rerank result
print('Calculating feature distances...')
target_features = target_features.numpy()
cluster = KMeans(n_clusters=numTarID, n_jobs=8, n_init=1)
# select & cluster images as training set of this epochs
print('Clustering and labeling...')
time0 = time.time()
clusterRes = cluster.fit(target_features)
time1 = time.time()
print('clustering cost', time1 - time0)
labels, centers = clusterRes.labels_, clusterRes.cluster_centers_
criterion[2] = ClassificationLoss(normalize(centers, axis=1))
# labels = splitLowconfi(target_features,labels,centers)
# num_ids = len(set(labels))
# print('Iteration {} have {} training ids'.format(iter_n+1, num_ids))
# generate new dataset
new_dataset = []
for (fname, _, cam, timestamp), label in zip(tgt_dataset.trainval, labels):
# if label==-1: continue
# dont need to change codes in trainer.py _parsing_input function and sampler function after add 0
new_dataset.append((fname,label,cam, timestamp))
print('Iteration {} have {} training images'.format(iter_n+1, len(new_dataset)))
train_loader = DataLoader(
Preprocessor(new_dataset, root=tgt_dataset.images_dir, transform=train_transformer),
batch_size=args.batch_size, num_workers=4,
sampler=RandomIdentitySampler(new_dataset, args.num_instances),
pin_memory=True, drop_last=True
)
evaluator = Evaluator(model, print_freq=args.print_freq)
if iter_n % 2 == 0:
trainer = Trainer(model, train_loader, criterion, optimizer)
for epoch in range(args.epochs):
trainer.train(epoch)
rank_score = evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
else:
classOptimizer = torch.optim.Adam([
{'params': model.parameters()},
{'params': criterion[2].classifier.parameters(), 'lr': 1e-3}
], lr=args.lr)
classTrainer = ClassTrainer(model, train_loader, criterion, classOptimizer)
for epoch in range(args.epochs):
# trainer.train(epoch, remRate=0.2+(0.6/args.iteration)*(1+iter_n)) # to at most 80%
classTrainer.train(epoch)
rank_score = evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
# Evaluate
rank_score = evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
save_checkpoint({
'state_dict': model.module.state_dict(),
'epoch': epoch + 1, 'best_top1': rank_score.market1501[0],
}, True, fpath=osp.join(args.logs_dir, 'adapted.pth.tar'))
return (rank_score.map, rank_score.market1501[0])
def get_data(data_dir,
source,
target,
height,
width,
batch_size,
re=0,
workers=8):
dataset = DA(data_dir, source, target)
normalizer = T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
num_classes = dataset.num_train_ids
train_transformer = T.Compose([
T.RandomSizedRectCrop(height, width),
T.RandomHorizontalFlip(),
T.ToTensor(),
normalizer,
T.RandomErasing(EPSILON=re),
])
test_transformer = T.Compose([
T.Resize((height, width), interpolation=3),
T.ToTensor(),
normalizer,
])
source_train_loader = DataLoader(Preprocessor(
dataset.source_train,
root=osp.join(dataset.source_images_dir, dataset.source_train_path),
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=True,
pin_memory=True,
drop_last=True)
target_train_loader = DataLoader(ModifiedTargetPreprocessor(
dataset.target_train,
root=osp.join(dataset.target_images_dir, dataset.target_train_path),
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=True,
pin_memory=True,
drop_last=True)
# target_train_loader = DataLoader(
# UnsupervisedCamStylePreprocessor(dataset.target_train,
# root=osp.join(dataset.target_images_dir, dataset.target_train_path),
# camstyle_root=osp.join(dataset.target_images_dir,
# dataset.target_train_camstyle_path),
# num_cam=dataset.target_num_cam, transform=train_transformer),
# batch_size=batch_size, num_workers=workers,
# shuffle=True, pin_memory=True, drop_last=True)
query_loader = DataLoader(Preprocessor(dataset.query,
root=osp.join(
dataset.target_images_dir,
dataset.query_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
gallery_loader = DataLoader(Preprocessor(dataset.gallery,
root=osp.join(
dataset.target_images_dir,
dataset.gallery_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
return dataset, num_classes, source_train_loader, target_train_loader, query_loader, gallery_loader
def get_data(dataname,
data_dir,
height,
width,
batch_size,
trainer,
re=0,
workers=8):
root = osp.join(data_dir, dataname)
dataset = datasets.create(dataname, root, trainer)
normalizer = T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_transformer = T.Compose([
T.RandomSizedRectCrop(height, width),
T.RandomHorizontalFlip(),
T.ToTensor(),
normalizer,
T.RandomErasing(EPSILON=re),
])
test_transformer = T.Compose([
T.Resize((height, width), interpolation=3),
T.ToTensor(),
normalizer,
])
if trainer <= 0:
num_classes = dataset.num_train_ids
train_loader = DataLoader(Preprocessor(dataset.train,
root=osp.join(
dataset.images_dir,
dataset.train_path),
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=True,
pin_memory=True,
drop_last=True)
query_loader = DataLoader(Preprocessor(dataset.query,
root=osp.join(
dataset.images_dir,
dataset.query_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
gallery_loader = DataLoader(Preprocessor(dataset.gallery,
root=osp.join(
dataset.images_dir,
dataset.gallery_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
return dataset, num_classes, train_loader, query_loader, gallery_loader
elif trainer == 1:
num_classes = dataset.num_pair_ids
pair_loader = DataLoader(Preprocessor_Pair(
dataset.pair,
root=osp.join(dataset.images_dir, dataset.pair_path),
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=True,
pin_memory=True,
drop_last=True)
query_loader = DataLoader(Preprocessor_Pair(
dataset.query,
root=osp.join(dataset.images_dir, dataset.query_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
gallery_loader = DataLoader(Preprocessor_Pair(
dataset.gallery,
root=osp.join(dataset.images_dir, dataset.gallery_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
return dataset, num_classes, pair_loader, query_loader, gallery_loader
elif trainer == 2:
num_classes = dataset.num_triplet_ids
triplet_loader = DataLoader(Preprocessor_Triplet(
dataset.triplet,
root=osp.join(dataset.images_dir, dataset.triplet_path),
transform=train_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=True,
pin_memory=True,
drop_last=True)
query_loader = DataLoader(Preprocessor_Pair(
dataset.query,
root=osp.join(dataset.images_dir, dataset.query_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
gallery_loader = DataLoader(Preprocessor_Pair(
dataset.gallery,
root=osp.join(dataset.images_dir, dataset.gallery_path),
transform=test_transformer),
batch_size=batch_size,
num_workers=workers,
shuffle=False,
pin_memory=True)
return dataset, num_classes, triplet_loader, query_loader, gallery_loader
def main(args):
np.random.seed(args.seed) # With the seed reset (every time), the same set of numbers will appear every time.
torch.manual_seed(args.seed) # Sets the seed for generating random numbers.
cudnn.benchmark = True # This flag allows you to enable the inbuilt cudnn auto-tuner to find the best algorithm to use for your hardware. It enables benchmark mode in cudnn.
# Redirect print to both console and log file
sys.stdout = Logger(osp.join(args.logs_dir, 'log.txt'))
print(args)
shutil.copy(sys.argv[0], osp.join(args.logs_dir,
osp.basename(sys.argv[0])))
# Create data loaders
if args.height is None or args.width is None:
args.height, args.width = (256, 128)
dataset, num_classes, train_loader, val_loader, test_loader = \
get_data(args.dataset, args.split, args.data_dir, args.height,
args.width, args.batch_size * 8, args.workers, #https://deeplizard.com/learn/video/kWVgvsejXsE#:~:text=The%20num_workers%20attribute%20tells%20the,sequentially%20inside%20the%20main%20process.
)
# Create model
model = models.create("ft_net_inter",
num_classes=num_classes, stride=args.stride)
# Load from checkpoint
start_epoch = 0
best_top1 = 0
top1 = 0
is_best = False
if args.checkpoint is not None:
if args.evaluate:
checkpoint = load_checkpoint(args.checkpoint)
param_dict = model.state_dict() # A state_dict is simply a Python dictionary object that maps each layer to its parameter tensor.
for k, v in checkpoint['state_dict'].items():
if 'model' in k:
param_dict[k] = v
model.load_state_dict(param_dict)
else:
model.model.load_param(args.checkpoint)
model = model.cuda()
# Distance metric
metric = None
# Evaluator
evaluator = Evaluator(model, use_cpu=args.use_cpu)
if args.evaluate:
print("Test:")
evaluator.evaluate(test_loader, dataset.query, dataset.gallery, metric)
return
train_transformer = [
T.Resize((args.height, args.width), interpolation=3),
T.RandomHorizontalFlip(),
T.Pad(10),
T.RandomCrop((args.height, args.width)),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
T.RandomErasing(probability=0.5),
]
train_transformer = T.Compose(train_transformer)
for cluster_epoch in range(args.cluster_epochs):
# -------------------------Stage 1 intra camera training--------------------------
# Cluster and generate new dataset and model
# Divides the training set into (subsets) and according to that each camera id is there for each image
# then it forms clustering on each subset according to the pair wise similarity
# then assigning images with in each cluster with identical label
# then cross entropy loss is used
cluster_result = get_intra_cam_cluster_result(model, train_loader,
args.class_number_stage1,
args.linkage)
cluster_datasets = [
datasets.create("cluster", osp.join(args.data_dir, args.dataset),
cluster_result[cam_id], cam_id)
for cam_id in cluster_result.keys()
]
cluster_dataloaders = [
DataLoader(Preprocessor(dataset.train_set,
root=dataset.images_dir,
transform=train_transformer),
batch_size=args.batch_size,
num_workers=args.workers,
shuffle=True,
pin_memory=False,
drop_last=True) for dataset in cluster_datasets
]
param_dict = model.model.state_dict()
model = models.create("ft_net_intra",
num_classes=[
args.class_number_stage1
for cam_id in cluster_result.keys()
],
stride=args.stride)
model_param_dict = model.model.state_dict()
for k, v in model_param_dict.items():
if k in param_dict.keys():
model_param_dict[k] = param_dict[k]
model.model.load_state_dict(model_param_dict)
model = model.cuda()
criterion = nn.CrossEntropyLoss().cuda()
# Optimizer
param_groups = make_params(model, args.lr, args.weight_decay)
optimizer = torch.optim.SGD(param_groups, momentum=0.9)
# Trainer
trainer = IntraCameraTrainer(
model, criterion, warm_up_epoch=args.warm_up)
print("start training")
# Start training
for epoch in range(0, args.epochs_stage1):
trainer.train(cluster_epoch,
epoch,
cluster_dataloaders,
optimizer,
print_freq=args.print_freq,
)
# -------------------------------------------Stage 2 inter camera training-----------------------------------
mix_rate = get_mix_rate(
args.mix_rate, cluster_epoch, args.cluster_epochs, power=args.decay_factor)
cluster_result = get_inter_cam_cluster_result(
model,
train_loader,
args.class_number_stage2,
args.linkage,
mix_rate,
use_cpu=args.use_cpu)
cluster_dataset = datasets.create(
"cluster", osp.join(args.data_dir, args.dataset), cluster_result,
0)
cluster_dataloaders = DataLoader(
Preprocessor(cluster_dataset.train_set,
root=cluster_dataset.images_dir,
transform=train_transformer),
batch_size=args.batch_size_stage2,
num_workers=args.workers,
sampler=RandomIdentitySampler(cluster_dataset.train_set,
args.batch_size_stage2,
args.instances),
pin_memory=False,
drop_last=True)
param_dict = model.model.state_dict()
model = models.create("ft_net_inter",
num_classes=args.class_number_stage2,
stride=args.stride)
model.model.load_state_dict(param_dict)
model = model.cuda()
# Criterion
criterion_entropy = nn.CrossEntropyLoss().cuda()
criterion_triple = TripletLoss(margin=args.margin).cuda()
# Optimizer
param_groups = make_params(model,
args.lr * args.batch_size_stage2 / 32,
args.weight_decay)
optimizer = torch.optim.SGD(param_groups, momentum=0.9)
# Trainer
trainer = InterCameraTrainer(model,
criterion_entropy,
criterion_triple,
warm_up_epoch=args.warm_up,
)
print("start training")
# Start training
for epoch in range(0, args.epochs_stage2):
trainer.train(cluster_epoch,
epoch,
cluster_dataloaders,
optimizer,
print_freq=args.print_freq)
if (cluster_epoch + 1) % 5 == 0: # in 4th, 9th, 14th epochs, see in the output
evaluator = Evaluator(model, use_cpu=args.use_cpu)
top1, mAP = evaluator.evaluate(
test_loader, dataset.query, dataset.gallery, metric, return_mAP=True)
is_best = top1 > best_top1
best_top1 = max(top1, best_top1)
save_checkpoint(
{
'state_dict': model.state_dict(),
'epoch': cluster_epoch + 1,
'best_top1': best_top1,
'cluster_epoch': cluster_epoch + 1,
},
is_best,
fpath=osp.join(args.logs_dir, 'checkpoint.pth.tar'))
if cluster_epoch == (args.cluster_epochs - 1):
save_checkpoint(
{
'state_dict': model.state_dict(),
'epoch': cluster_epoch + 1,
'best_top1': best_top1,
'cluster_epoch': cluster_epoch + 1,
},
False,
fpath=osp.join(args.logs_dir, 'latest.pth.tar'))
print('\n * cluster_epoch: {:3d} top1: {:5.1%} best: {:5.1%}{}\n'.
format(cluster_epoch, top1, best_top1, ' *' if is_best else ''))
# Final test
print('Test with best model:')
checkpoint = load_checkpoint(osp.join(args.logs_dir, 'model_best.pth.tar'))
model.load_state_dict(checkpoint['state_dict'])
best_rank1, mAP = evaluator.evaluate(
test_loader, dataset.query, dataset.gallery, metric, return_mAP=True)
def main(args):
np.random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.benchmark = True
# Create data loaders
assert args.num_instances > 1, "num_instances should be greater than 1"
assert args.batch_size % args.num_instances == 0, \
'num_instances should divide batch_size'
if args.height is None or args.width is None:
args.height, args.width = (144, 56) if args.arch == 'inception' else \
(256, 128)
# get source data
src_dataset, src_extfeat_loader = \
get_source_data(args.src_dataset, args.data_dir, args.height,
args.width, args.batch_size, args.workers)
# get target data
tgt_dataset, num_classes, tgt_extfeat_loader, test_loader = \
get_data(args.tgt_dataset, args.data_dir, args.height,
args.width, args.batch_size, args.workers)
# Create model
# Hacking here to let the classifier be the number of source ids
if args.src_dataset == 'dukemtmc':
model = models.create(args.arch, num_classes=632, pretrained=False)
elif args.src_dataset == 'market1501':
model = models.create(args.arch, num_classes=676, pretrained=False)
else:
raise RuntimeError(
'Please specify the number of classes (ids) of the network.')
# Load from checkpoint
if args.resume:
print(
'Resuming checkpoints from finetuned model on another dataset...\n'
)
checkpoint = load_checkpoint(args.resume)
model.load_state_dict(checkpoint['state_dict'], strict=False)
else:
raise RuntimeWarning('Not using a pre-trained model.')
model = nn.DataParallel(model).cuda()
# evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
# if args.evaluate: return
# Criterion
criterion = [SortedTripletLoss(args.margin, isAvg=True).cuda()]
# Optimizer
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
# training stage transformer on input images
normalizer = T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_transformer = T.Compose([
T.Resize((args.height, args.width)),
T.RandomHorizontalFlip(),
T.ToTensor(), normalizer,
T.RandomErasing(probability=0.5, sh=0.2, r1=0.3)
])
evaluator = Evaluator(model, print_freq=args.print_freq)
evaluator.evaluate(test_loader, tgt_dataset.query, tgt_dataset.gallery)
st_model = ST_Model(tgt_dataset.meta['num_cameras'])
# # Start training
for iter_n in range(args.iteration):
if args.lambda_value == 0:
source_features = 0
else:
# get source datas' feature
source_features, _ = extract_features(model,
src_extfeat_loader,
print_freq=args.print_freq)
# synchronization feature order with src_dataset.train
source_features = torch.cat([
source_features[f].unsqueeze(0)
for f, _, _, _ in src_dataset.train
], 0)
# extract training images' features
print('Iteration {}: Extracting Target Dataset Features...'.format(
iter_n + 1))
target_features, tarNames = extract_features(
model, tgt_extfeat_loader, print_freq=args.print_freq)
# synchronization feature order with dataset.train
target_features = torch.cat([
target_features[f].unsqueeze(0)
for f, _, _, _ in tgt_dataset.trainval
], 0)
# target_real_label = np.asarray([tarNames[f].unsqueeze(0) for f, _, _, _ in tgt_dataset.trainval])
# calculate distance and rerank result
target_features = target_features.numpy()
rerank_dist = re_ranking(source_features,
target_features,
lambda_value=args.lambda_value)
# if iter_n > 0:
# rerank_dist = st_model.apply(rerank_dist, tgt_dataset.trainval, tgt_dataset.trainval)
cluster = HDBSCAN(metric='precomputed', min_samples=10)
# select & cluster images as training set of this epochs
clusterRes = cluster.fit(rerank_dist)
labels, label_num = clusterRes.labels_, clusterRes.labels_.max() + 1
# centers = np.zeros((label_num, target_features.shape[1]))
# nums = [0] * target_features.shape[1]
print('clusters num =', label_num)
# generate new dataset
new_dataset = []
index = -1
for (fname, _, cam, timestamp), label in zip(tgt_dataset.trainval,
labels):
index += 1
if label == -1: continue
# dont need to change codes in trainer.py _parsing_input function and sampler function after add 0
new_dataset.append((fname, label, cam, timestamp))
# centers[label] += target_features[index]
# nums[label] += 1
print('Iteration {} have {} training images'.format(
iter_n + 1, len(new_dataset)))
# learn ST model
same, _ = st_model.fit(new_dataset)
# st_model.fit(tgt_dataset.trainval)
def filter(i, j):
_, _, c1, t1 = tgt_dataset.trainval[i]
_, _, c2, t2 = tgt_dataset.trainval[j]
return same.in_peak(c1, c2, t1, t2, 0.2)
ranking = np.argsort(rerank_dist)[:, 1:]
cluster_size = 23.535612535612536
must_conn = int(cluster_size / 2)
might_conn = int(cluster_size * 2)
length = len(tgt_dataset.trainval)
pos = [[] for _ in range(length)]
neg = [[] for _ in range(length)]
for i in range(length):
for j_ in range(might_conn):
j = ranking[i][j_]
if j_ < must_conn and i in ranking[j][:must_conn]:
pos[i].append(j)
elif i in ranking[j][:might_conn] and filter(i, j):
pos[i].append(j)
# if j_ < must_conn or filter(i, j):
# pos[i].append(j)
else:
neg[i].append(j)
if len(neg[i]) < len(pos[i]):
neg[i].extend(ranking[i][j_ + 1:j_ + 1 + len(pos[i]) -
len(neg[i])])
# learn visual model
# for i in range(label_num):
# centers[i] /= nums[i]
# criterion[3] = ClassificationLoss(normalize(centers, axis=1)).cuda()
#
# classOptimizer = torch.optim.Adam([
# {'params': model.parameters()},
# {'params': criterion[3].classifier.parameters(), 'lr': 1e-3}
# ], lr=args.lr)
train_loader = DataLoader(Preprocessor(tgt_dataset.trainval,
root=tgt_dataset.images_dir,
transform=train_transformer),
batch_size=args.batch_size,
num_workers=4,
sampler=TripletSampler(
tgt_dataset.trainval, pos, neg),
pin_memory=True,
drop_last=True)
trainer = Trainer(model, train_loader, criterion, optimizer)
for epoch in range(args.epochs):
trainer.train(epoch)
rank_score = evaluator.evaluate(test_loader, tgt_dataset.query,
tgt_dataset.gallery)
# Evaluate
rank_score = evaluator.evaluate(test_loader, tgt_dataset.query,
tgt_dataset.gallery)
save_checkpoint(
{
'state_dict': model.module.state_dict(),
'epoch': epoch + 1,
'best_top1': rank_score.market1501[0],
},
True,
fpath=osp.join(args.logs_dir, 'adapted.pth.tar'))
return (rank_score.map, rank_score.market1501[0])