def get_net(args):
load_pretrained = args.net_init == 'pretrained' and args.checkpoint == ""
if load_pretrained:
print(yellow('Loading a net, pretrained on ImageNet1k.'))
model = polynet(num_classes=1000,
pretrained='imagenet' if load_pretrained else None)
feature_extractor = nn.Sequential(
model.stem,
model.stage_a,
model.reduction_a,
model.stage_b,
model.reduction_b,
model.stage_c,
)
num_classes = [int(x) for x in args.num_classes.split(',')]
# Predictor
predictor = MultiHead(in_features=model.last_linear.in_features,
num_classes=num_classes)
predictor = nn.Sequential(nn.Dropout(args.dropout_p), predictor)
# Construct
model = BaseModel(feature_extractor, predictor)
return model
def get_net(args):
load_pretrained = args.net_init == 'pretrained' and args.checkpoint == ""
if load_pretrained:
print(yellow('Loading a net, pretrained on ImageNet1k.'))
model = sys.modules[__name__].__dict__[args.arch](
num_classes=1000, pretrained='imagenet' if load_pretrained else None)
num_classes = [int(x) for x in args.num_classes.split(',')]
# Extractor
feature_extractor = nn.Sequential(model.layer0, model.layer1, model.layer2,
model.layer3, model.layer4)
# Predictor
predictor = MultiHead(in_features=model.last_linear.in_features,
num_classes=num_classes)
# if args.dropout_p > 0:
predictor = nn.Sequential(nn.Dropout(args.dropout_p), predictor)
# Construct
model = BaseModel(feature_extractor, predictor)
return model
def get_net(args):
load_pretrained = args.net_init == 'pretrained' and args.checkpoint == ""
if load_pretrained:
print(yellow('Loading a net, pretrained on ImageNet1k.'))
model = SE_ResNeXt101FT(num_classes=340, pretrained=load_pretrained)
if not args.load_only_extractor:
return model
model.load_state_dict(
torch.load(
'extensions/qd/data/experiments/se_resnext_n01z3/checkpoints/se_resnext101_n.pth'
)['state_dict'])
# Extractor
feature_extractor = model.features
# Construct
num_classes = [int(x) for x in args.num_classes.split(',')]
predictor = nn.Sequential(
nn.Dropout(args.dropout_p),
MultiHead(in_features=2048, num_classes=num_classes))
model = BaseModel(feature_extractor, predictor)
return model
def get_net(args):
load_pretrained = args.net_init == 'pretrained' and args.checkpoint == ""
if load_pretrained:
print(yellow('Loading a net, pretrained on ImageNet1k.'))
model = se_resnext50(num_classes=1000, pretrained=load_pretrained)
# if not args.load_only_extractor:
# return model
# model.load_state_dict(torch.load('extensions/qd/data/experiments/se_resnext_n01z3/checkpoints/se_resnext101_n.pth')['state_dict'])
if args.num_input_channels != 3:
# if args.num_input_channels % 3 != 0:
# assert False
conv1_ = model.conv1
model.conv1 = torch.nn.Conv2d(args.num_input_channels,
conv1_.out_channels,
kernel_size=conv1_.kernel_size,
stride=conv1_.stride,
padding=conv1_.padding,
bias=False)
for i in range(int(args.num_input_channels / 3)):
model.conv1.weight.data[:, i * 3:(i + 1) *
3] = conv1_.weight.data / (int(
args.num_input_channels / 3))
if args.num_input_channels % 3 > 0:
model.conv1.weight.data[:, -(args.num_input_channels %
3):] = conv1_.weight.data[:, -(
args.num_input_channels % 3):]
feature_extractor = nn.Sequential(model.conv1, model.bn1, model.relu,
model.maxpool, model.layer1,
model.layer2, model.layer3, model.layer4)
num_classes = [int(x) for x in args.num_classes.split(',')]
predictor = nn.Sequential(
nn.Dropout(args.dropout_p),
MultiHead(in_features=2048, num_classes=num_classes))
model = BaseModel(feature_extractor, predictor)
return model
def get_net(args):
load_pretrained = args.net_init == 'pretrained' and args.checkpoint == ""
if load_pretrained:
print(yellow('Loading a net, pretrained on ImageNet1k.'))
feature_extractor = ResidualNet('ImageNet',
args.depth,
1000,
args.att_type,
pretrained=load_pretrained)
num_classes = [int(x) for x in args.num_classes.split(',')]
predictor = MultiHead(in_features=feature_extractor.fc.in_features,
num_classes=num_classes)
predictor = nn.Sequential(nn.Dropout(args.dropout_p), predictor)
feature_extractor.fc = None
model = BaseModel(feature_extractor, predictor)
return model
def get_net(args):
load_pretrained = args.net_init == 'pretrained' and args.checkpoint == ""
if load_pretrained:
print(yellow('Loading a net, pretrained on ImageNet1k.'))
model = inceptionresnetv2(num_classes=1000, pretrained='imagenet' if load_pretrained else None)
feature_extractor = nn.Sequential(
model.conv2d_1a,
model.conv2d_2a,
model.conv2d_2b,
model.maxpool_3a,
model.conv2d_3b,
model.conv2d_4a,
model.maxpool_5a,
model.mixed_5b,
model.repeat,
model.mixed_6a,
model.repeat_1,
model.mixed_7a,
model.repeat_2,
model.block8,
model.conv2d_7b,
)
num_classes = [int(x) for x in args.num_classes.split(',')]
# Predictor
predictor = MultiHead(in_features = model.last_linear.in_features, num_classes=num_classes)
predictor = nn.Sequential( nn.Dropout(args.dropout_p), predictor)
# Construct
model = BaseModel(feature_extractor, predictor)
return model
trainloader = DataLoader(dataset=trainset,
batch_size=args.batch_size,
shuffle=(train_sampler is None),
sampler=train_sampler,
num_workers=args.num_workers,
pin_memory=True)
valloader = DataLoader(dataset=valset,
batch_size=args.batch_size,
shuffle=False,
num_workers=args.num_workers,
pin_memory=True)
# model
model = BaseModel(model_name=args.model_name,
num_classes=args.num_classes,
pretrained=args.pretrained)
if args.resume:
state = torch.load(args.resume)
print('Resume from:{}'.format(args.resume))
model.load_state_dict(state['net'], strict=False)
best_acc = state['acc']
start_epoch = state['epoch'] + 1
if 'ols' in args.loss:
criterion['ols'].matrix = state['ols'].cuda()
# sync_bn
if args.sync_bn and multi_gpus:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
print('Using SyncBatchNorm')
def drop(db=DATABASE_URI):
"""Drop the tables."""
connect(db)
BaseModel.drop_tables()
def create(db=DATABASE_URI):
"""Create the tables."""
connect(db)
BaseModel.create_tables()
mode = 'best'
parser = argparse.ArgumentParser()
parser.add_argument('--savepath', default='./Base224L2/eff-b3', type=str)
parser.add_argument('--last', action='store_true')
args = parser.parse_args()
path = args.savepath
if args.last:
mode = 'last'
args = get_setting(path)
# print(args)
# model
model = BaseModel(model_name=args['model_name'], num_classes=int(args['num_classes']), \
pretrained=int(args['pretrained']), pool_type=args['pool_type'], down=int(args['down']))
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# device = torch.device('cpu')
model = model.to(device)
load_pretrained_model(path, model, mode=mode)
# data
testset = Testset(root='./data/test')
testloader = DataLoader(dataset=testset,
batch_size=128,
shuffle=False,
num_workers=8,
pin_memory=True)
submit = {}
