def __init__(self, args):
self.args = args
# image transform
input_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize([.485, .456, .406], [.229, .224, .225]),
])
# dataset and dataloader
data_kwargs = {'transform': input_transform,
'base_size': args.base_size,
'crop_size': args.crop_size}
train_dataset = get_segmentation_dataset(args.dataset, split=args.train_split, mode='train', **data_kwargs)
val_dataset = get_segmentation_dataset(args.dataset, split='val', mode='val', **data_kwargs)
self.train_loader = data.DataLoader(dataset=train_dataset,
batch_size=args.batch_size,
shuffle=True,
drop_last=True)
self.val_loader = data.DataLoader(dataset=val_dataset,
batch_size=1,
shuffle=False)
# create network
self.model = get_fast_scnn(dataset=args.dataset, aux=args.aux)
if torch.cuda.device_count() > 1:
self.model = torch.nn.DataParallel(self.model, device_ids=[0, 1, 2])
self.model.to(args.device)
# resume checkpoint if needed
if args.resume:
if os.path.isfile(args.resume):
name, ext = os.path.splitext(args.resume)
assert ext == '.pkl' or '.pth', 'Sorry only .pth and .pkl files supported.'
print('Resuming training, loading {}...'.format(args.resume))
self.model.load_state_dict(torch.load(args.resume, map_location=lambda storage, loc: storage))
# create criterion
self.criterion = MixSoftmaxCrossEntropyOHEMLoss(aux=args.aux, aux_weight=args.aux_weight,
ignore_index=-1).to(args.device)
# optimizer
self.optimizer = torch.optim.SGD(self.model.parameters(),
lr=args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
# lr scheduling
self.lr_scheduler = LRScheduler(mode='poly', base_lr=args.lr, nepochs=args.epochs,
iters_per_epoch=len(self.train_loader),
power=0.9)
# evaluation metrics
self.metric = SegmentationMetric(train_dataset.num_class)
self.best_pred = 0.0
def __init__(self, args, cfg=None):
# train_dataset = [build_dataset(cfg.data.train)]
# self.dataset= train_dataset
# val_dataset = [build_dataset(cfg.data.test)]
# if len(cfg.workflow) == 2:
# train_dataset.append(build_dataset(cfg.data.val))
# train_data_loaders = [
# build_dataloader(
# ds,
# cfg.data.imgs_per_gpu,
# cfg.data.workers_per_gpu,
# # cfg.gpus,
# dist=False) for ds in train_dataset
# ]
# val_data_loader = build_dataloader(
# val_dataset,
# imgs_per_gpu=1,
# workers_per_gpu=cfg.data.workers_per_gpu,
# dist=False,
# shuffle=False)
# self.train_loader = train_data_loaders[0]
# self.val_loader = val_data_loader
self.args = args
# image transform
input_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375]),
])
# dataset and dataloader
data_kwargs = {
'transform': input_transform,
'base_size': args.base_size,
'crop_size': args.crop_size
}
train_dataset = get_segmentation_dataset(args.dataset,
split=args.train_split,
mode='train',
**data_kwargs)
val_dataset = get_segmentation_dataset(args.dataset,
split='val',
mode='val',
**data_kwargs)
self.train_loader = data.DataLoader(dataset=train_dataset,
batch_size=args.batch_size,
shuffle=True,
drop_last=True)
self.val_loader = data.DataLoader(dataset=val_dataset,
batch_size=1,
shuffle=False)
# create network
self.model = get_fast_scnn(dataset=args.dataset, aux=args.aux)
if torch.cuda.device_count() > 1:
self.model = torch.nn.DataParallel(self.model,
device_ids=[0, 1, 2])
self.model.to(args.device)
# resume checkpoint if needed
if args.resume:
if os.path.isfile(args.resume):
name, ext = os.path.splitext(args.resume)
assert ext == '.pkl' or '.pth', 'Sorry only .pth and .pkl files supported.'
print('Resuming training, loading {}...'.format(args.resume))
self.model.load_state_dict(
torch.load(args.resume,
map_location=lambda storage, loc: storage))
# create criterion
self.criterion = MixSoftmaxCrossEntropyOHEMLoss(
aux=args.aux, aux_weight=args.aux_weight,
ignore_index=-1).to(args.device)
# optimizer
self.optimizer = torch.optim.SGD(self.model.parameters(),
lr=args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
# lr scheduling
self.lr_scheduler = LRScheduler(mode='poly',
base_lr=args.lr,
nepochs=args.epochs,
iters_per_epoch=len(self.train_loader),
power=0.9)
# evaluation metrics
self.metric = SegmentationMetric(train_dataset.num_class)
self.best_pred = 0.0