def train(optimizer, num_classes, num_epochs, scheduler, device):
load = get_dataset()
model = get_model_instance_segmentation(num_classes)
model = model.to(device)
if optimizer == 'Adam':
exp_optimizer = optim.Adam(model.parameters(), lr=1e-3)
else:
exp_optimizer = optim.SGD(model.parameters(),
lr=0.005,
momentum=0.9,
weight_decay=0.0005)
if scheduler:
lr_scheduler = optim.lr_scheduler.StepLR(exp_optimizer,
step_size=3,
gamma=0.1)
for epoch in range(num_epochs):
train_one_epoch(model,
exp_optimizer,
load['train'],
device,
epoch,
print_freq=10)
lr_scheduler.step()
evaluate(model, load['val'], device=device)
torch.save(model.state_dict(), 'best_model')
print('Finished')
def train(data_root):
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
# background and person
num_classes = 2
dataset = PennFudanDataset(data_root,get_transform(train=True))
dataset_test = PennFudanDataset(data_root,get_transform(train=False))
# split the dataset
indices = torch.randperm(len(dataset)).tolist()
dataset = Subset(dataset,indices[:-50])
dataset_test = Subset(dataset_test,indices[-50:])
# define data loaders
data_loader = DataLoader(dataset, batch_size=2, shuffle=True, num_workers=4,
collate_fn=tools.collate_fn)
data_loader_test = DataLoader(dataset_test, batch_size=1, shuffle=False, num_workers=4,
collate_fn=tools.collate_fn)
# get model
model = get_model_instance_segmentation(num_classes)
model.to(device)
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.SGD(params,lr=0.005,momentum=0.9,weight_decay=0.0005)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer,step_size=3)
num_epochs =10
for epoch in range(num_epochs):
train_one_epoch(model,optimizer,data_loader,device,epoch,print_freq=10)
lr_scheduler.step()
# evaluate(model,data_loader_test,device=device)
torch.save(model.state_dict(),"masknet.pth")
print("OK!")
'loss_bb_regression_max', 'loss_bb_regression_min', 'loss_classifier_avg',
'loss_classifier_median', 'loss_classifier_max', 'loss_classifier_min',
'loss_rpn_bb_regression_avg', 'loss_rpn_bb_regression_median',
'loss_rpn_bb_regression_max', 'loss_rpn_bb_regression_min'
]
train_epochs_log = pd.DataFrame(columns=columns_epochs)
train_iterations_log = pd.DataFrame(columns=columns_iterations)
# Train the network (saving the best model)
for epoch in range(0, epochs):
# train for one epoch, printing every <print_freq> iterations
training_results, train_iterations_log = train_one_epoch(
model,
optimizer,
loader_train,
device,
epoch,
print_freq=1,
df=train_iterations_log)
# add epoch logs to df
train_epochs_log = helper.df_add_epoch_log(train_epochs_log, epoch,
training_results)
# evaluate on the validation data set
mAP = evaluate(model, loader_validation, device=device)
# Check to keep best model
if mAP > best_mAP:
best_mAP = mAP
# Save model
# Training loop
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.SGD(params,
lr=0.005,
momentum=0.9,
weight_decay=0.0005)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=3,
gamma=0.1)
num_epochs = 10
for e in range(num_epochs):
# train for one epoch
train_one_epoch(model,
optimizer,
train_loader,
device,
e,
print_freq=10)
# update learning rate
lr_scheduler.step()
# evaluate on the test dataset
print('entering eval')
print(len(val_loader))
evaluate(model, val_loader, device=device)
# save model
if e % 10 == 0:
torch.save({
'epoch': e,
'model_state_dict': model.state_dict()
}, f'leaf_od' + str(e) + 'EPOCH_checkpoint.pt')
def main(args):
utils.init_distributed_mode(args)
print(args)
device = torch.device(args.device)
# Data loading code
print("Loading data")
dataset, num_classes = get_dataset(args.dataset, "train",
get_transform(train=True),
args.data_path)
dataset_test, _ = get_dataset(args.dataset, "val",
get_transform(train=False), args.data_path)
print("Creating data loaders")
if args.distributed:
train_sampler = torch.utils.data.distributed.DistributedSampler(
dataset)
test_sampler = torch.utils.data.distributed.DistributedSampler(
dataset_test)
else:
train_sampler = torch.utils.data.RandomSampler(dataset)
test_sampler = torch.utils.data.SequentialSampler(dataset_test)
if args.aspect_ratio_group_factor >= 0:
group_ids = create_aspect_ratio_groups(
dataset, k=args.aspect_ratio_group_factor)
train_batch_sampler = GroupedBatchSampler(train_sampler, group_ids,
args.batch_size)
else:
train_batch_sampler = torch.utils.data.BatchSampler(train_sampler,
args.batch_size,
drop_last=True)
data_loader = torch.utils.data.DataLoader(
dataset,
batch_sampler=train_batch_sampler,
num_workers=args.workers,
collate_fn=utils.collate_fn)
data_loader_test = torch.utils.data.DataLoader(dataset_test,
batch_size=1,
sampler=test_sampler,
num_workers=args.workers,
collate_fn=utils.collate_fn)
print("Creating model")
# model = torchvision.models.detection.__dict__[args.model](num_classes=num_classes,
# pretrained=args.pretrained)
model = get_model(num_classes=num_classes)
model.to(device)
model_without_ddp = model
if args.distributed:
model = torch.nn.parallel.DistributedDataParallel(
model, device_ids=[args.gpu])
model_without_ddp = model.module
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.SGD(params,
lr=args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
# lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=args.lr_step_size, gamma=args.lr_gamma)
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer, milestones=args.lr_steps, gamma=args.lr_gamma)
if args.resume:
print("----------------------Resume--------------")
checkpoint = torch.load(args.resume, map_location='cpu')
model_without_ddp.load_state_dict(checkpoint['model'])
optimizer.load_state_dict(checkpoint['optimizer'])
lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
args.start_epoch = checkpoint['epoch'] + 1
if args.test_only:
evaluate(model, data_loader_test, device=device)
return
print("Start training")
start_time = time.time()
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
train_sampler.set_epoch(epoch)
train_one_epoch(model, optimizer, data_loader, device, epoch,
args.print_freq)
lr_scheduler.step()
if args.output_dir:
utils.save_on_master(
{
'model': model_without_ddp.state_dict(),
'optimizer': optimizer.state_dict(),
'lr_scheduler': lr_scheduler.state_dict(),
'args': args,
'epoch': epoch
}, os.path.join(args.output_dir, 'model_{}.pth'.format(epoch)))
# evaluate after every epoch
evaluate(model, data_loader_test, device=device)
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
print('Training time {}'.format(total_time_str))
def main(args):
input_size = (224, 224)
best_acc = 0.0
# prepare output folder
if args.output_dir:
if not Path(args.output_dir).is_dir():
Path(args.output_dir).mkdir()
# read config
with open(args.cfg, 'r') as f:
cfg_dict = yaml.load(f, Loader=yaml.FullLoader)
config_stem = Path(args.cfg).stem
hyp = cfg_dict['hyp']
data = cfg_dict['data']
names = np.unique(
data['names']
) # sort as sklearn.preprocessing.LabelEncoder.fit_transform() does
# set device mode
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# create model
model_name = args.model
nc = data['nc']
feature_extract = hyp['feature_extract']
print('[INFO] Creating model ({})'.format(model_name))
model, input_size = initialize_model(model_name, nc, feature_extract)
model.to(device)
# load data
print('[INFO] Loading data')
train_csv = data['train']
val_csv = data['val']
train_dataset, val_dataset, train_sampler = load_data_from_csv(
train_csv, val_csv, input_size, args.transform)
# dataloader
batch_size = hyp['batch_size']
train_loader = DataLoader(train_dataset,
batch_size=batch_size,
sampler=train_sampler,
num_workers=args.workers)
val_loader = DataLoader(val_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=args.workers)
# criterion + optimizer + scheduler
learning_rate = hyp['lr']
momentum = hyp['momentum']
weight_decay = hyp['weight_decay']
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(),
lr=learning_rate,
momentum=momentum,
weight_decay=weight_decay)
# scheduler = optim.lr_scheduler.MultiStepLR(optimizer,milestones=[0.5*args.total_epochs, 0.8*args.total_epochs], gamma=0.1)
# create tensorboard writter
logdir = f'runs/{model_name}_{config_stem}'
writter = SummaryWriter(log_dir=logdir)
if args.resume:
print('[INFO] Load checkpoint')
ckpt = torch.load(args.resume, map_location=device)
model.load_state_dict(ckpt['model_state_dict'])
optimizer.load_state_dict(ckpt['optimizer'])
args.start_epoch = ckpt['epoch'] + 1
best_acc = ckpt['best_acc'] if 'best_acc' in ckpt else ckpt['acc']
if args.eval:
ckpt_ = torch.load(args.eval, map_location=device)
model.load_state_dict(ckpt_['model'])
evaluate(val_loader, model, names, device)
return
# train
start_epoch = args.start_epoch
total_epochs = hyp['total_epochs']
try:
print('[INFO] Starting training')
start_time = time.time()
for epoch in range(start_epoch, total_epochs):
epoch_info = f'Epoch {epoch}/{total_epochs-1}'
print(epoch_info)
print('-' * len(epoch_info))
# train engine
train_acc, train_loss = train_one_epoch(train_loader, model,
criterion, optimizer,
epoch, device)
val_acc, val_loss = validate(val_loader, model, criterion, device)
# scheduler.step()
# logging to tensorboard
writter.add_scalar('Loss/train', train_loss, epoch)
writter.add_scalar('Loss/val', val_loss, epoch)
writter.add_scalar('Acc/train', train_acc, epoch)
writter.add_scalar('Acc/val', val_acc, epoch)
# print training info
info = f'loss ' + f'{train_loss:.3f} ' + f'accuracy ' + f'{train_acc:.1f}% ' \
+ f'val_loss ' + f'{val_loss:.3f} ' + f'val_accuracy ' + f'{val_acc:.1f}%' + '\n'
print(info)
is_best = val_acc > best_acc
if is_best:
best_acc = val_acc
print('Found new best val_acc: {:6.2f}!\n'.format(best_acc))
# save checkpoint each 10 epochs
checkpoint = {
'epoch': epoch,
'acc': val_acc,
'model': model,
'model_state_dict': model.state_dict(),
'optimizer': optimizer.state_dict()
}
filepath = str(
Path(args.output_dir).joinpath(
f'{model_name}_{config_stem}.pt'))
save_checkpoint(checkpoint, filepath, epoch, is_best)
except KeyboardInterrupt:
print('[INFO] Training interrupted. Saving checkpoint')
print('[INFO] Best val_acc: {:.2f}'.format(best_acc))
filepath = str(
Path(args.output_dir).joinpath(
f'{model_name}_{config_stem}_{epoch-1}.pt'))
save_checkpoint(checkpoint, filepath, epoch, force_save=True)
writter.flush()
writter.close()
sys.exit(0)
# flush and close tensorboard writter
writter.flush()
writter.close()
elapsed_time = time.time() - start_time
elapsed_str = str(datetime.timedelta(seconds=int(elapsed_time)))
print('[INFO] Training complete in: {}'.format(elapsed_str))
print('[INFO] Best val_acc: {:.2f}'.format(best_acc))
filepath = str(
Path(args.output_dir).joinpath(f'{model_name}_{config_stem}_final.pt'))
save_checkpoint(checkpoint, filepath, epoch, force_save=True)
weight_decay=0.0005)
# and a learning rate scheduler which decreases the learning rate by
# 10x every 3 epochs
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=15,
gamma=0.1)
# TRAINING LOOP
save_fr = 1
print_freq = 25 # make sure that print_freq is smaller than len(dataset) & len(dataset_test)
os.makedirs('./maskrcnn_saved_models', exist_ok=True)
for epoch in range(num_epochs):
# train for one epoch, printing every 10 iterations
train_one_epoch(model,
optimizer,
data_loader,
device,
epoch,
print_freq=print_freq)
if epoch % save_fr == 0:
torch.save(
model.state_dict(),
'./maskrcnn_saved_models/mask_rcnn_model_epoch_{}.pt'.format(
str(epoch)))
# update the learning rate
lr_scheduler.step()
# evaluate on the test dataset
evaluate(model, data_loader_test, device=device)