def run(settings):
# Most common settings are assigned in the settings struct
settings.description = 'distilled ATOM IoUNet with default settings according to the paper.'
settings.batch_size = 32
settings.num_workers = 8
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 5.0
settings.feature_sz = 18
settings.output_sz = settings.feature_sz * 16
settings.center_jitter_factor = {'train': 0, 'test': 4.5}
settings.scale_jitter_factor = {'train': 0, 'test': 0.5}
# Train datasets
lasot_train = Lasot(settings.env.lasot_dir, split='train')
trackingnet_train = TrackingNet(settings.env.trackingnet_dir,
set_ids=list(range(11)))
coco_train = MSCOCOSeq(settings.env.coco_dir)
# Validation datasets
trackingnet_val = TrackingNet(settings.env.trackingnet_dir,
set_ids=list(range(11, 12)))
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
# The augmentation transform applied to the training set (individually to each image in the pair)
transform_train = tfm.Transform(
tfm.ToTensorAndJitter(0.2),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# The augmentation transform applied to the validation set (individually to each image in the pair)
transform_val = tfm.Transform(
tfm.ToTensor(),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# Data processing to do on the training pairs
proposal_params = {
'min_iou': 0.1,
'boxes_per_frame': 16,
'sigma_factor': [0.01, 0.05, 0.1, 0.2, 0.3]
}
data_processing_train = processing.ATOMProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_train,
joint_transform=transform_joint)
# Data processing to do on the validation pairs
data_processing_val = processing.ATOMProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_val,
joint_transform=transform_joint)
# The sampler for training
dataset_train = sampler.ATOMSampler(
[lasot_train, trackingnet_train, coco_train], [1, 1, 1],
samples_per_epoch=1000 * settings.batch_size,
max_gap=50,
processing=data_processing_train)
# The loader for training
loader_train = LTRLoader('train',
dataset_train,
training=True,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=True,
drop_last=True,
stack_dim=1)
# The sampler for validation
dataset_val = sampler.ATOMSampler([trackingnet_val], [1],
samples_per_epoch=500 *
settings.batch_size,
max_gap=50,
processing=data_processing_val)
# The loader for validation
loader_val = LTRLoader('val',
dataset_val,
training=False,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=False,
drop_last=True,
epoch_interval=5,
stack_dim=1)
# Load teacher network
teacher_net = atom_models.atom_resnet18(backbone_pretrained=True)
teacher_path = '/home/ddanier/CFKD/pytracking/networks/atom_default.pth'
teacher_net = loading.load_weights(teacher_net, teacher_path, strict=True)
print(
'*******************Teacher net loaded successfully*******************'
)
# Create student network and actor
student_net = atom_models.atom_resnet18small(backbone_pretrained=False)
objective = distillation.CFKDLoss(
reg_loss=nn.MSELoss(),
w_cf=0.01,
w_fd=100,
cf_layers=['conv1', 'layer1', 'layer2', 'layer3'])
actor = actors.AtomCompressionActor(student_net, teacher_net, objective)
# Optimizer
optimizer = optim.Adam(
[{
'params': actor.student_net.feature_extractor.parameters()
}, {
'params': actor.student_net.bb_regressor.parameters()
}],
lr=1e-2)
lr_scheduler = optim.lr_scheduler.StepLR(optimizer,
step_size=15,
gamma=0.1)
# Create trainer
trainer = LTRDistillationTrainer(actor, [loader_train, loader_val],
optimizer, settings, lr_scheduler)
# Run training (set fail_safe=False if you are debugging)
trainer.train(50, load_latest=False, fail_safe=True)
def run(settings):
# Most common settings are assigned in the settings struct
settings.description = 'ATOM IoUNet with default settings, but additionally using GOT10k for training.'
settings.batch_size = 64
settings.num_workers = 8
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 5.0
settings.feature_sz = 18
settings.output_sz = settings.feature_sz * 16
settings.center_jitter_factor = {'train': 0, 'test': 4.5}
settings.scale_jitter_factor = {'train': 0, 'test': 0.5}
# Train datasets
# lasot_train = Lasot(settings.env.lasot_dir, split='train')
# got10k_train = Got10k(settings.env.got10k_dir, split='vottrain')
# trackingnet_train = TrackingNet(settings.env.trackingnet_dir, set_ids=list(range(4)))
# coco_train = MSCOCOSeq(settings.env.coco_dir)
coco_train = MSCOCOSeq_depth(settings.env.cocodepth_dir,
dtype='rgbcolormap')
lasot_depth_train = Lasot_depth(root=settings.env.lasotdepth_dir,
dtype='rgbcolormap')
depthtrack_train = DepthTrack(root=settings.env.depthtrack_dir,
dtype='rgbcolormap')
depthtrack_horizontal_train = DepthTrack(
root=settings.env.depthtrack_horizontal_dir, dtype='rgbcolormap')
depthtrack_vertical_train = DepthTrack(
root=settings.env.depthtrack_vertical_dir, dtype='rgbcolormap')
# Validation datasets
# got10k_val = Got10k(settings.env.got10k_dir, split='votval')
cdtb_val = CDTB(settings.env.cdtb_dir, split='val', dtype='rgbcolormap')
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
# The augmentation transform applied to the training set (individually to each image in the pair)
transform_train = tfm.Transform(
tfm.ToTensorAndJitter(0.2),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# The augmentation transform applied to the validation set (individually to each image in the pair)
transform_val = tfm.Transform(
tfm.ToTensor(),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# Data processing to do on the training pairs
proposal_params = {
'min_iou': 0.1,
'boxes_per_frame': 16,
'sigma_factor': [0.01, 0.05, 0.1, 0.2, 0.3]
}
data_processing_train = processing.ATOMProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_train,
joint_transform=transform_joint)
# Data processing to do on the validation pairs
data_processing_val = processing.ATOMProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_val,
joint_transform=transform_joint)
# The sampler for training
dataset_train = sampler.ATOMSampler([
lasot_depth_train, depthtrack_train, depthtrack_horizontal_train,
depthtrack_vertical_train, coco_train
], [1, 1, 1, 1, 1],
samples_per_epoch=1000 *
settings.batch_size,
max_gap=50,
processing=data_processing_train)
# The loader for training
loader_train = LTRLoader('train',
dataset_train,
training=True,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=True,
drop_last=True,
stack_dim=1)
# The sampler for validation
dataset_val = sampler.ATOMSampler([cdtb_val], [1],
samples_per_epoch=500 *
settings.batch_size,
max_gap=50,
processing=data_processing_val)
# The loader for validation
loader_val = LTRLoader('val',
dataset_val,
training=False,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=False,
drop_last=True,
epoch_interval=5,
stack_dim=1)
# Create network and actor
net = atom_models.atom_resnet18_DeT(
backbone_pretrained=True,
merge_type='max') # 'mean', 'conv', 'weightedSum'
objective = nn.MSELoss()
actor = actors.AtomActor(net=net, objective=objective)
# Optimizer
# optimizer = optim.Adam(actor.net.bb_regressor.parameters(), lr=1e-3)
optimizer = optim.Adam(
[
{
'params': actor.net.bb_regressor.parameters()
},
{
'params': actor.net.feature_extractor.parameters(),
'lr': 2e-5
},
{
'params': actor.net.feature_extractor_depth.parameters(),
'lr': 2e-5
},
# {'params': actor.net.merge_layer2.parameters(), 'lr': 2e-5},
# {'params': actor.net.merge_layer3.parameters(), 'lr': 2e-5},
],
lr=1e-3)
lr_scheduler = optim.lr_scheduler.StepLR(optimizer,
step_size=15,
gamma=0.2)
# Create trainer
trainer = LTRTrainer(actor, [loader_train, loader_val], optimizer,
settings, lr_scheduler)
# Run training (set fail_safe=False if you are debugging)
trainer.train(80, load_latest=True, fail_safe=True)
def run(settings):
# Most common settings are assigned in the settings struct
settings.device = 'cuda'
settings.description = 'TransT with default settings.'
settings.batch_size = 38
settings.num_workers = 8
settings.multi_gpu = True
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 4.0
settings.template_area_factor = 2.0
settings.search_feature_sz = 32
settings.template_feature_sz = 16
settings.search_sz = settings.search_feature_sz * 8
settings.temp_sz = settings.template_feature_sz * 8
settings.center_jitter_factor = {'search': 3, 'template': 0}
settings.scale_jitter_factor = {'search': 0.25, 'template': 0}
# Transformer
settings.position_embedding = 'sine'
settings.hidden_dim = 256
settings.dropout = 0.1
settings.nheads = 8
settings.dim_feedforward = 2048
settings.featurefusion_layers = 4
# Train datasets
lasot_train = Lasot(settings.env.lasot_dir, split='train')
got10k_train = Got10k(settings.env.got10k_dir, split='vottrain')
trackingnet_train = TrackingNet(settings.env.trackingnet_dir, set_ids=list(range(4)))
coco_train = MSCOCOSeq(settings.env.coco_dir)
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
# The augmentation transform applied to the training set (individually to each image in the pair)
transform_train = tfm.Transform(tfm.ToTensorAndJitter(0.2),
tfm.Normalize(mean=settings.normalize_mean, std=settings.normalize_std))
# Data processing to do on the training pairs
data_processing_train = processing.TransTProcessing(search_area_factor=settings.search_area_factor,
template_area_factor = settings.template_area_factor,
search_sz=settings.search_sz,
temp_sz=settings.temp_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
transform=transform_train,
joint_transform=transform_joint)
# The sampler for training
dataset_train = sampler.TransTSampler([lasot_train, got10k_train, coco_train, trackingnet_train], [1,1,1,1],
samples_per_epoch=1000*settings.batch_size, max_gap=100, processing=data_processing_train)
# The loader for training
loader_train = LTRLoader('train', dataset_train, training=True, batch_size=settings.batch_size, num_workers=settings.num_workers,
shuffle=True, drop_last=True, stack_dim=0)
# Create network and actor
model = transt_models.transt_resnet50(settings)
# Wrap the network for multi GPU training
if settings.multi_gpu:
model = MultiGPU(model, dim=0)
objective = transt_models.transt_loss(settings)
n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('number of params:', n_parameters)
actor = actors.TranstActor(net=model, objective=objective)
# Optimizer
param_dicts = [
{"params": [p for n, p in model.named_parameters() if "backbone" not in n and p.requires_grad]},
{
"params": [p for n, p in model.named_parameters() if "backbone" in n and p.requires_grad],
"lr": 1e-5,
},
]
optimizer = torch.optim.AdamW(param_dicts, lr=1e-4,
weight_decay=1e-4)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, 500)
# Create trainer
trainer = LTRTrainer(actor, [loader_train], optimizer, settings, lr_scheduler)
# Run training (set fail_safe=False if you are debugging)
trainer.train(1000, load_latest=True, fail_safe=True)
def run(settings):
# Most common settings are assigned in the settings struct
settings.description = 'construct training settings for depth-based rgbd visual tracking'
settings.batch_size = 2
settings.num_workers = 2
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 5.0
settings.feature_sz = 18
settings.output_sz = settings.feature_sz * 16
settings.center_jitter_factor = {'train': 0, 'test': 4.5}
settings.scale_jitter_factor = {'train': 0, 'test': 0.5}
# Train datasets
cdtb_train = CDTB()
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
# The augmentation transform applied to the training set (individually to each image in the pair)
transform_train = tfm.Transform(tfm.ToTensorAndJitter(0.2),
tfm.Normalize(mean=settings.normalize_mean, std=settings.normalize_std))
# The augmentation transform applied to the validation set (individually to each image in the pair)
transform_val = tfm.Transform(tfm.ToTensor(),
tfm.Normalize(mean=settings.normalize_mean, std=settings.normalize_std))
# Data processing to do on the training pairs
proposal_params = {'min_iou': 0.1, 'boxes_per_frame': 16, 'sigma_factor': [0.01, 0.05, 0.1, 0.2, 0.3]}
data_processing_train = processing.DepthProcessing(search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_train,
joint_transform=transform_joint)
# Data processing to do on the validation pairs
data_processing_val = processing.DepthProcessing(search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_val,
joint_transform=transform_joint)
# The sampler for training
dataset_train = sampler.DepthSampler([cdtb_train], p_datasets=None,
samples_per_epoch=1000*settings.batch_size, max_gap=50, processing=data_processing_train)
# The loader for training
loader_train = LTRLoader('train', dataset_train, training=True, batch_size=settings.batch_size, num_workers=settings.num_workers,
shuffle=True, drop_last=True, stack_dim=1)
# The sampler for validation
dataset_val = sampler.DepthSampler([cdtb_train], [1], samples_per_epoch=500*settings.batch_size, max_gap=50,
processing=data_processing_val)
# The loader for validation
loader_val = LTRLoader('val', dataset_val, training=False, batch_size=settings.batch_size, num_workers=settings.num_workers,
shuffle=False, drop_last=True, epoch_interval=5, stack_dim=1)
# Create network and actor
net = depth_models.depth_atom_resnet50(backbone_pretrained=True)
objective = nn.MSELoss()
actor = actors.AtomActor(net=net, objective=objective)
# Optimizer
optimizer = optim.Adam(actor.net.bb_regressor.parameters(), lr=1e-3)
lr_scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=15, gamma=0.2)
# Create trainer
trainer = LTRTrainer(actor, [loader_train, loader_val], optimizer, settings, lr_scheduler)
# Run training (set fail_safe=False if you are debugging)
trainer.train(50, load_latest=True, fail_safe=True)
def run(settings):
# Most common settings are assigned in the settings struct
debug = False
if debug:
settings.batch_size = 8 # 8 # 4 # 120 # 70 # 38
settings.num_workers = 0 # 24 # 30 # 10 # 35 # 30 min(settings.batch_size, 16)
settings.multi_gpu = False # True # True # True # True # True # True # True
else:
settings.batch_size = 64 # 8 # 4 # 120 # 70 # 38
settings.num_workers = 20 # 24 # 30 # 10 # 35 # 30 min(settings.batch_size, 16)
settings.multi_gpu = True # True # True # True # True # True # True # True
settings.device = 'cuda'
settings.description = 'TransT with default settings.'
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 4.0
settings.template_area_factor = 2.
settings.search_feature_sz = 32
settings.template_feature_sz = 16
settings.search_sz = settings.search_feature_sz * 8
settings.temp_sz = settings.template_feature_sz * 8
settings.center_jitter_factor = {'search': 3.0, 'template': 0} # 3
settings.scale_jitter_factor = {'search': 0.25, 'template': 0} # 0.25
settings.sequence_length = 8 # 30 # 64 NEXT # Same as PT
settings.rand = True
# settings.search_gap = 1 # Depreciated
settings.init_ckpt = "pytracking/networks/transt.pth"
# Transformer
settings.position_embedding = 'sine'
settings.hidden_dim = 256
settings.dropout = 0.1
settings.nheads = 8
settings.dim_feedforward = 2048
settings.featurefusion_layers = 4
# settings.sigma = 1 / 4 / 5.
# settings.kernel = 4
# settings.feature = 32 # 18
# settings.output_sz = 256 # settings.feature * 16
# settings.end_pad_if_even = False
# settings.label_function_params = True
settings.move_data_to_gpu = True
# Train datasets
lasot_train = Lasot(settings.env.lasot_dir, split='train')
got10k_train = Got10k(settings.env.got10k_dir, split='vottrain') # votval
trackingnet_train = TrackingNet(settings.env.trackingnet_dir,
set_ids=list(range(4)))
# coco_train = MSCOCOSeq(settings.env.coco_dir)
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
# The augmentation transform applied to the training set (individually to each image in the pair)
# transform_train = tfm.Transform(tfm.ToTensorAndJitter(0.2),
# tfm.Normalize(mean=settings.normalize_mean, std=settings.normalize_std))
transform_train = tfm.Transform(
tfm.ToTensorAndJitter(0.2),
# tfm.RandomHorizontalFlip(),
# tfm.RandomAffine(p_flip=0.5, max_scale=1.5),
# tfm.RandomBlur(1),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# Data processing to do on the training pairs
data_processing_train = processing.TransTProcessing(
search_area_factor=settings.search_area_factor,
template_area_factor=settings.template_area_factor,
search_sz=settings.search_sz,
temp_sz=settings.temp_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
joint=False, # Whether or not to apply same transform to every image
transform=transform_train,
rand=settings.rand,
label_function_params=None, # settings.label_function_params,
joint_transform=transform_joint)
# The sampler for training
# dataset_train = sampler.TransTSampler([got10k_train], [1], samples_per_epoch=1000*settings.batch_size, max_gap=100, processing=data_processing_train, num_search_frames=settings.sequence_length, frame_sample_mode="rnn_causal")
dataset_train = sampler.TransTSampler(
[lasot_train, got10k_train, trackingnet_train], [1, 1, 1],
samples_per_epoch=1000 * settings.batch_size,
max_gap=settings.sequence_length * 4,
processing=data_processing_train,
num_search_frames=settings.sequence_length,
frame_sample_mode="rnn_interval")
# dataset_train = sampler.TransTSampler([lasot_train, got10k_train, trackingnet_train], [1,1,1], samples_per_epoch=1000*settings.batch_size, max_gap=100, processing=data_processing_train)
# The loader for training
loader_train = LTRLoader('train',
dataset_train,
training=True,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=True,
drop_last=True,
stack_dim=0,
pin_memory=settings.move_data_to_gpu == False)
# Create network and actor
model = transt_models.transt_resnet50(settings)
# Wrap the network for multi GPU training
if settings.multi_gpu:
model = MultiGPU(model, dim=0)
objective = transt_models.transt_loss(settings)
n_parameters = sum(p.numel() for p in model.parameters()
if p.requires_grad)
print('number of params:', n_parameters)
actor = actors.OldCircuitTranstActor(net=model, objective=objective)
# Optimizer
# Change learning rate forthe Q we have changed and the RNN and the readout
# q = self.mix_q(torch.cat([q, self.mix_norm(exc)], -1))
# self.class_embed_new = MLP(hidden_dim * 2, hidden_dim, num_classes + 1, 3)
# self.bbox_embed_new = MLP(hidden_dim * 2, hidden_dim, 4, 3)
param_dicts = [
# {"params": [p for n, p in model.named_parameters() if "backbone" not in n and p.requires_grad and "circuit" not in n and "mix" not in n and "new" not in n]},
# {
# "params": [p for n, p in model.named_parameters() if "backbone" in n and p.requires_grad],
# "lr": 1e-5,
# },
{
"params": [
p for n, p in model.named_parameters()
if "encoder" not in n or "input_proj" not in n or "decoder"
not in n or "backbone" not in n or "embed" not in n
], # "exc" in n or "circuit" in n or "mix" in n or "new" in n or "rnn" in n or "step" in n], # or "class_embed" in n or "bbox_embed" in n],
"lr":
1e-4,
},
# {
# "params": [p for n, p in model.named_parameters() if "class_embed" in n or "bbox_embed" in n], # if "circuit" in n or "mix" in n or "new" in n or "class_embed" in n or "bbox_embed" in n],
# "lr": 1e-5,
# },
]
print("Higher lr:")
print([
n for n, p in model.named_parameters()
if "encoder" not in n and "input_proj" not in n and "decoder" not in n
and "backbone" not in n and "embed" not in n
])
print("*" * 60)
for n, p in model.named_parameters():
# if "circuit" in n or "mix" in n or "new" in n or "class_embed" in n or "bbox_embed" in n or "decoder" in n or "encoder" in n or "rnn" in n:
if "backbone" not in n: # "circuit" in n or "exc" in n or "mix" or "class_embed" in n or "bbox_embed" in n or "new" in n or "rnn" in n or "encoder" in n or "decoder" in n or "step" in n or "proj" in n: # or "decoder" in n: # "class_embed" in n or "bbox_embed" in n or "rnn" in n:
# if "encoder" not in n or "input_proj" not in n or "decoder" not in n or "backbone" not in n or "embed" not in n:
print("TRAINING {}".format(n))
else:
p.requires_grad = False
print("Removing grad on {}".format(n))
optimizer = torch.optim.AdamW(
param_dicts,
lr=1e-4, # lr=1e-4,
weight_decay=1e-4)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, 500)
# Create trainer
trainer = LTRTrainer(actor, [loader_train], optimizer, settings,
lr_scheduler)
# Run training (set fail_safe=False if you are debugging)
trainer.train(1000, load_latest=True, fail_safe=True)
def run(settings):
settings.description = 'Default train settings for DiMP with ResNet50 as backbone.'
settings.batch_size = 10
settings.num_workers = 8
settings.multi_gpu = False
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 5.0
settings.output_sigma_factor = 1 / 4
settings.target_filter_sz = 4
settings.feature_sz = 18
settings.output_sz = settings.feature_sz * 16
settings.center_jitter_factor = {'train': 3, 'test': 4.5}
settings.scale_jitter_factor = {'train': 0.25, 'test': 0.5}
settings.hinge_threshold = 0.05
# settings.print_stats = ['Loss/total', 'Loss/iou', 'ClfTrain/clf_ce', 'ClfTrain/test_loss']
# Train datasets
# lasot_train = Lasot(settings.env.lasot_dir, split='train')
# got10k_train = Got10k(settings.env.got10k_dir, split='vottrain')
# trackingnet_train = TrackingNet(settings.env.trackingnet_dir, set_ids=list(range(4)))
# coco_train = MSCOCOSeq(settings.env.coco_dir)
dtype = 'colormap'
cdtb_depth_train = CDTB(root=settings.env.cdtb_dir,
split=None,
dtype=dtype)
cdtb_st_train = CDTB_ST(root=settings.env.cdtb_st_dir,
split=None,
dtype=dtype)
# depthtrack_train = DepthTrack(root=settings.env.depthtrack_dir, split='train', dtype=dtype)
coco_train = MSCOCOSeq_depth(settings.env.cocodepth_dir, dtype=dtype)
got10k_depth_train = Got10k_depth(settings.env.got10kdepth_dir,
dtype=dtype)
lasot_depth_train = Lasot_depth(root=settings.env.lasotdepth_dir,
rgb_root=settings.env.lasot_dir,
dtype=dtype)
# depthtrack_horizontal_train = DepthTrack(root=settings.env.depthtrack_horizontal_dir, split='train', dtype='color')
# depthtrack_vertical_train = DepthTrack(root=settings.env.depthtrack_vertical_dir, split='train', dtype='color')
# Validation datasets
# got10k_val = Got10k(settings.env.got10k_dir, split='votval')
depthtrack_val = DepthTrack(root=settings.env.depthtrack_dir,
split='val',
dtype=dtype)
# Data transform
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
transform_train = tfm.Transform(
tfm.ToTensorAndJitter(0.2),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
transform_val = tfm.Transform(
tfm.ToTensor(),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# The tracking pairs processing module
output_sigma = settings.output_sigma_factor / settings.search_area_factor
proposal_params = {
'min_iou': 0.1,
'boxes_per_frame': 8,
'sigma_factor': [0.01, 0.05, 0.1, 0.2, 0.3]
}
label_params = {
'feature_sz': settings.feature_sz,
'sigma_factor': output_sigma,
'kernel_sz': settings.target_filter_sz
}
data_processing_train = processing.DiMPProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
label_function_params=label_params,
transform=transform_train,
joint_transform=transform_joint)
data_processing_val = processing.DiMPProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
label_function_params=label_params,
transform=transform_val,
joint_transform=transform_joint)
# Train sampler and loader
# dataset_train = sampler.DiMPSampler([lasot_train, got10k_train, trackingnet_train, coco_train], [0.25,1,1,1],
# samples_per_epoch=26000, max_gap=30, num_test_frames=3, num_train_frames=3,
# processing=data_processing_train)
dataset_train = sampler.DiMPSampler([cdtb_depth_train], [1],
samples_per_epoch=26000,
max_gap=30,
num_test_frames=3,
num_train_frames=3,
processing=data_processing_train)
loader_train = LTRLoader('train',
dataset_train,
training=True,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=True,
drop_last=True,
stack_dim=1)
# Validation samplers and loaders
# dataset_val = sampler.DiMPSampler([got10k_val], [1], samples_per_epoch=5000, max_gap=30,
# num_test_frames=3, num_train_frames=3,
# processing=data_processing_val)
dataset_val = sampler.DiMPSampler([depthtrack_val], [1],
samples_per_epoch=5000,
max_gap=30,
num_test_frames=3,
num_train_frames=3,
processing=data_processing_val)
loader_val = LTRLoader('val',
dataset_val,
training=False,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=False,
drop_last=True,
epoch_interval=5,
stack_dim=1)
# Create network and actor
net = dimpnet.dimpnet50(filter_size=settings.target_filter_sz,
backbone_pretrained=True,
optim_iter=5,
clf_feat_norm=True,
clf_feat_blocks=0,
final_conv=True,
out_feature_dim=512,
optim_init_step=0.9,
optim_init_reg=0.1,
init_gauss_sigma=output_sigma *
settings.feature_sz,
num_dist_bins=100,
bin_displacement=0.1,
mask_init_factor=3.0,
target_mask_act='sigmoid',
score_act='relu')
# Wrap the network for multi GPU training
if settings.multi_gpu:
net = MultiGPU(net, dim=1)
objective = {
'iou': nn.MSELoss(),
'test_clf': ltr_losses.LBHinge(threshold=settings.hinge_threshold)
}
loss_weight = {
'iou': 1,
'test_clf': 100,
'test_init_clf': 100,
'test_iter_clf': 400
}
actor = actors.DiMPActor(net=net,
objective=objective,
loss_weight=loss_weight)
# Optimizer
optimizer = optim.Adam(
[{
'params': actor.net.classifier.filter_initializer.parameters(),
'lr': 5e-5
}, {
'params': actor.net.classifier.filter_optimizer.parameters(),
'lr': 5e-4
}, {
'params': actor.net.classifier.feature_extractor.parameters(),
'lr': 5e-5
}, {
'params': actor.net.bb_regressor.parameters()
}, {
'params': actor.net.feature_extractor.parameters(),
'lr': 2e-5
}],
lr=2e-4)
# optimizer = optim.Adam([{'params': actor.net.classifier.filter_initializer.parameters(), 'lr': 1e-4},
# {'params': actor.net.classifier.filter_optimizer.parameters(), 'lr': 1e-4},
# {'params': actor.net.classifier.feature_extractor.parameters(), 'lr': 1e-45},
# {'params': actor.net.bb_regressor.parameters()},
# {'params': actor.net.feature_extractor.parameters(), 'lr': 1e-4}],
# lr=1e-4)
lr_scheduler = optim.lr_scheduler.StepLR(optimizer,
step_size=15,
gamma=0.2)
# lr_scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=50, gamma=0.2)
trainer = LTRTrainer(actor, [loader_train, loader_val], optimizer,
settings, lr_scheduler)
trainer.train(150, load_latest=True, fail_safe=True)
def run(settings):
# Most common settings are assigned in the settings struct
settings.description = 'SiamFC with Alexnet backbone and trained with vid'
settings.print_interval = 100 # How often to print loss and other info
settings.batch_size = 8 # Batch size
settings.num_workers = 8 # Number of workers for image loading
settings.normalize_mean = [0., 0., 0.] # Normalize mean
settings.normalize_std = [1 / 255., 1 / 255., 1 / 255.] # Normalize std
settings.search_area_factor = {
'train': 1.0,
'test': 2.0078740157480315
} # roughly the same as SiamFC
settings.output_sz = {'train': 127, 'test': 255}
settings.scale_type = 'context'
settings.border_type = 'meanpad'
# Settings for the image sample and proposal generation
settings.center_jitter_factor = {'train': 0, 'test': 0}
settings.scale_jitter_factor = {'train': 0, 'test': 0.}
# Train datasets
vid_train = ImagenetVID()
# Validation datasets
got10k_val = vid_train #Got10k(split='val')
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = dltransforms.ToGrayscale(probability=0.25)
# The augmentation transform applied to the training set (individually to each image in the pair)
transform_exemplar = dltransforms.Compose([
dltransforms.ToArray(),
dltransforms.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std)
])
transform_instance = dltransforms.Compose([
DataAug(),
dltransforms.ToArray(),
dltransforms.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std)
])
# Data processing to do on the training pairs
data_processing_train = processing.SiamFCProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
scale_type=settings.scale_type,
border_type=settings.border_type,
mode='sequence',
train_transform=transform_exemplar,
test_transform=transform_instance,
joint_transform=transform_joint)
# Data processing to do on the validation pairs
data_processing_val = processing.SiamFCProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
scale_type=settings.scale_type,
border_type=settings.border_type,
mode='sequence',
transform=transform_exemplar,
joint_transform=transform_joint)
# The sampler for training
dataset_train = sampler.ATOMSampler([vid_train], [
1,
],
samples_per_epoch=6650 *
settings.batch_size,
max_gap=100,
processing=data_processing_train)
# The loader for training
train_loader = loader.LTRLoader('train',
dataset_train,
training=True,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
stack_dim=1)
# The sampler for validation
dataset_val = sampler.ATOMSampler([got10k_val], [
1,
],
samples_per_epoch=1000 *
settings.batch_size,
max_gap=100,
processing=data_processing_val)
# The loader for validation
val_loader = loader.LTRLoader('val',
dataset_val,
training=False,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
epoch_interval=5,
stack_dim=1)
# creat network, set objective, creat optimizer, learning rate scheduler, trainer
with dygraph.guard():
# Create network
net = siamfc_alexnet()
# Create actor, which wraps network and objective
actor = actors.SiamFCActor(net=net,
objective=None,
batch_size=settings.batch_size,
shape=(17, 17),
radius=16,
stride=8)
# Set to training mode
actor.train()
# define optimizer and learning rate
lr_scheduler = fluid.layers.exponential_decay(learning_rate=0.01,
decay_steps=6650,
decay_rate=0.8685,
staircase=True)
regularizer = fluid.regularizer.L2DecayRegularizer(
regularization_coeff=0.0005)
optimizer = fluid.optimizer.Momentum(momentum=0.9,
regularization=regularizer,
parameter_list=net.parameters(),
learning_rate=lr_scheduler)
trainer = LTRTrainer(actor, [train_loader], optimizer, settings,
lr_scheduler)
trainer.train(50, load_latest=False, fail_safe=False)
def run(settings):
# Most common settings are assigned in the settings struct
settings.description = 'ATOM using the probabilistic maximum likelihood trained regression model for bounding-box' \
'regression presented in [https://arxiv.org/abs/1909.12297].'
settings.batch_size = 64
settings.num_workers = 8
settings.print_interval = 1
settings.normalize_mean = [0.485, 0.456, 0.406]
settings.normalize_std = [0.229, 0.224, 0.225]
settings.search_area_factor = 5.0
settings.feature_sz = 18
settings.output_sz = settings.feature_sz * 16
settings.center_jitter_factor = {'train': 0, 'test': 4.5}
settings.scale_jitter_factor = {'train': 0, 'test': 0.5}
# Train datasets
lasot_train = Lasot(settings.env.lasot_dir, split='train')
got10k_train = Got10k(settings.env.got10k_dir, split='vottrain')
trackingnet_train = TrackingNet(settings.env.trackingnet_dir,
set_ids=list(range(4)))
coco_train = MSCOCOSeq(settings.env.coco_dir)
# Validation datasets
got10k_val = Got10k(settings.env.got10k_dir, split='votval')
# The joint augmentation transform, that is applied to the pairs jointly
transform_joint = tfm.Transform(tfm.ToGrayscale(probability=0.05))
# The augmentation transform applied to the training set (individually to each image in the pair)
transform_train = tfm.Transform(
tfm.ToTensorAndJitter(0.2),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# The augmentation transform applied to the validation set (individually to each image in the pair)
transform_val = tfm.Transform(
tfm.ToTensor(),
tfm.Normalize(mean=settings.normalize_mean,
std=settings.normalize_std))
# Data processing to do on the training pairs
proposal_params = {
'boxes_per_frame': 128,
'gt_sigma': (0, 0),
'proposal_sigma': [(0.05, 0.05), (0.5, 0.5)],
'add_mean_box': True
}
data_processing_train = processing.KLBBregProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_train,
joint_transform=transform_joint)
# Data processing to do on the validation pairs
data_processing_val = processing.KLBBregProcessing(
search_area_factor=settings.search_area_factor,
output_sz=settings.output_sz,
center_jitter_factor=settings.center_jitter_factor,
scale_jitter_factor=settings.scale_jitter_factor,
mode='sequence',
proposal_params=proposal_params,
transform=transform_val,
joint_transform=transform_joint)
# The sampler for training
dataset_train = sampler.ATOMSampler(
[lasot_train, got10k_train, trackingnet_train, coco_train],
[1, 1, 1, 1],
samples_per_epoch=1000 * settings.batch_size,
max_gap=200,
processing=data_processing_train)
# The loader for training
loader_train = LTRLoader('train',
dataset_train,
training=True,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=True,
drop_last=True,
stack_dim=1)
# The sampler for validation
dataset_val = sampler.ATOMSampler([got10k_val], [1],
samples_per_epoch=500 *
settings.batch_size,
max_gap=200,
processing=data_processing_val)
# The loader for validation
loader_val = LTRLoader('val',
dataset_val,
training=False,
batch_size=settings.batch_size,
num_workers=settings.num_workers,
shuffle=False,
drop_last=True,
epoch_interval=5,
stack_dim=1)
# Create network and actor
net = atom_models.atom_resnet18(backbone_pretrained=True)
objective = klreg_losses.MLRegression()
actor = bbreg_actors.AtomBBKLActor(net=net, objective=objective)
# Optimizer
optimizer = optim.Adam(actor.net.bb_regressor.parameters(), lr=1e-3)
lr_scheduler = optim.lr_scheduler.StepLR(optimizer,
step_size=15,
gamma=0.2)
# Create trainer
trainer = LTRTrainer(actor, [loader_train, loader_val], optimizer,
settings, lr_scheduler)
# Run training (set fail_safe=False if you are debugging)
trainer.train(50, load_latest=True, fail_safe=True)
