parser.set_defaults(plot=False)
parser.add_argument('--checkpoint_directory',
type=str,
default='checkpoints',
help='directory for training checkpoints')
parser.add_argument('--checkpoint_prefix',
type=str,
default='rord',
help='prefix for training checkpoints')
args = parser.parse_args()
print(args)
# Creating CNN model
model = D2Net(model_file=args.init_model, use_cuda=False)
model = model.to(device)
# Optimizer
optimizer = optim.Adam(filter(lambda p: p.requires_grad, model.parameters()),
lr=args.lr)
training_dataset = PhotoTourismIPR(base_path=args.dataset_path,
preprocessing=args.preprocessing)
training_dataset.build_dataset()
training_dataloader = DataLoader(training_dataset,
batch_size=args.batch_size,
num_workers=args.num_workers)
help='prefix for training checkpoints')
args = parser.parse_args()
print(args)
# Create the folders for plotting if need be
if args.plot:
plot_path = 'train_vis'
if os.path.isdir(plot_path):
print('[Warning] Plotting directory already exists.')
else:
os.mkdir(plot_path)
# Creating CNN model
model = D2Net(model_file=args.model_file, use_cuda=use_cuda)
# Optimizer
optimizer = optim.Adam(filter(lambda p: p.requires_grad, model.parameters()),
lr=args.lr)
# Dataset
if args.use_validation:
validation_dataset = MegaDepthDataset(
scene_list_path='megadepth_utils/valid_scenes.txt',
scene_info_path=args.scene_info_path,
base_path=args.dataset_path,
train=False,
preprocessing=args.preprocessing,
pairs_per_scene=25)
validation_dataloader = DataLoader(validation_dataset,
parser.set_defaults(plot=False)
parser.add_argument('--checkpoint_directory',
type=str,
default='checkpoints',
help='directory for training checkpoints')
parser.add_argument('--checkpoint_prefix',
type=str,
default='d2',
help='prefix for training checkpoints')
args = parser.parse_args()
print(args)
# Creating CNN model
model = D2Net(model_file=args.model_file, use_cuda=False)
model = model.to(device)
# Optimizer
optimizer = optim.Adam(filter(lambda p: p.requires_grad, model.parameters()),
lr=args.lr)
# Dataset
if args.use_validation:
validation_dataset = PhotoTourismCombined(
# scene_list_path='megadepth_utils/valid_scenes.txt',
# scene_info_path=args.scene_info_path,
base_path=args.dataset_path,
train=False,
preprocessing=args.preprocessing,
pairs_per_scene=25)
import torch
from lib.model import D2Net
from lib.utils import preprocess_image
from lib.pyramid import process_multiscale
import scipy
import scipy.io
import scipy.misc
import numpy as np
use_cuda = torch.cuda.is_available()
# Creating CNN model
model = D2Net(model_file="models/d2_tf.pth", use_relu=True, use_cuda=use_cuda)
#현지가 임의로 수정하려다가 주석
device = torch.device("cuda:0" if use_cuda else "cpu")
#device = torch.device("cpu")
multiscale = True
max_edge = 2500
max_sum_edges = 5000
# de-net feature extract function
def cnn_feature_extract(image, scales=[.25, 0.50, 1.0], nfeatures=1000):
if len(image.shape) == 2:
image = image[:, :, np.newaxis]
image = np.repeat(image, 3, -1)
# TODO: switch to PIL.Image due to deprecation of scipy.misc.imresize.
resized_image = image
model_path = os.path.join(log_folder, 'model')
if not os.path.exists(log_path):
os.makedirs(log_path)
if not os.path.exists(model_path):
os.makedirs(model_path)
create_exp_dir(log_folder, scripts_to_save=glob.glob('*.py'))
create_exp_dir(log_folder, scripts_to_save=glob.glob('lib/*.py'))
vis_writer = SummaryWriter(log_path)
else:
vis_writer = None
criterion = D2Loss_hpatches(scaling_steps=3, device=device).to(device)
model = D2Net(model_file=args.resume, use_cuda=use_cuda)
# Optimizer
optimizer = optim.Adam(
filter(lambda p: p.requires_grad, model.parameters()), lr=args.lr
)
print(gct(), 'Param size = %fMB', count_parameters_in_MB(model))
# Resume training
if args.resume:
if os.path.isfile(args.resume):
checkpoint = torch.load(args.resume)
start_epoch = checkpoint['epoch'] + 1
optimizer.load_state_dict(checkpoint['optimizer'])
best_loss = checkpoint['loss']
scaling_steps = 6 - args.dilation_blocks - args.truncated_blocks
safe_radius = 2**(args.truncated_blocks + args.dilation_blocks - 1)
elif args.model_type == 'res101':
scaling_steps = 6 - args.dilation_blocks - args.truncated_blocks
safe_radius = 2**(args.truncated_blocks + args.dilation_blocks - 1)
# print information for check
print(
'The %s model is used for this testing (truncated at -%s block and finetune last %s layers) '
% (args.model_type, args.truncated_blocks, args.finetune_layers))
# Creating CNN model
model = D2Net(model_file=args.model_file,
use_cuda=use_cuda,
finetune_layers=args.finetune_layers,
truncated_blocks=args.truncated_blocks,
model_type=args.model_type,
finetune_skip_layers=args.finetune_skip_layers,
dilation_blocks=args.dilation_blocks)
# Optimizer
optimizer = optim.Adam(filter(lambda p: p.requires_grad, model.parameters()),
lr=args.lr)
# Dataset
if args.use_validation:
validation_dataset = MegaDepthDataset(
scene_list_path='megadepth_utils/valid_scenes.txt',
scene_info_path=args.scene_info_path,
base_path=args.dataset_path,
train=False,