print('Restored model, epoch {}, step {}'.format(epoch, step))
else:
print("No trained model detected")
raise OSError
# Validation
feature_descriptor_model.eval()
feature_descriptor_model = feature_descriptor_model.module
feature_descriptor_model = feature_descriptor_model.cuda(gpu_id)
# Summary network architecture
if display_architecture:
torchsummary.summary(feature_descriptor_model,
input_size=(3, height, width))
total_query = 0
folder_list = utils.get_parent_folder_names(testing_data_root,
id_range=testing_patient_id)
mean_accuracy_1 = None
mean_accuracy_2 = None
mean_accuracy_3 = None
for patient_id in testing_patient_id:
data_root = Path(testing_data_root) / "{:d}".format(patient_id)
sub_folders = list(data_root.glob("*/"))
sub_folders.sort()
for folder in sub_folders:
# Get color image filenames
test_filenames = utils.get_file_names_in_sequence(
sequence_root=folder)
if len(test_filenames) == 0:
print("Sequence {} does not have relevant files".format(
currentDT.month,
currentDT.day,
currentDT.hour,
currentDT.minute,
testing_patient_id)
if not log_root.exists():
log_root.mkdir()
writer = SummaryWriter(logdir=str(log_root))
print("Tensorboard visualization at {}".format(str(log_root)))
# Get color image filenames
train_filenames, val_filenames, test_filenames = utils.get_color_file_names_by_bag(training_data_root,
validation_patient_id=validation_patient_id,
testing_patient_id=testing_patient_id,
id_range=id_range)
folder_list = utils.get_parent_folder_names(training_data_root,
id_range=id_range)
# Build training and validation dataset
train_dataset = dataset.SfMDataset(image_file_names=train_filenames,
folder_list=folder_list,
adjacent_range=adjacent_range, transform=training_transforms,
downsampling=input_downsampling,
network_downsampling=network_downsampling, inlier_percentage=inlier_percentage,
use_store_data=load_intermediate_data,
store_data_root=training_data_root,
phase="train", is_hsv=is_hsv,
num_pre_workers=num_workers, visible_interval=30, rgb_mode="rgb")
validation_dataset = dataset.SfMDataset(image_file_names=val_filenames,
folder_list=folder_list,
adjacent_range=adjacent_range,
transform=None,
model_root = root / "models"
try:
model_root.mkdir(mode=0o777, parents=True)
except OSError:
pass
results_root = root / "results"
try:
results_root.mkdir(mode=0o777, parents=True)
except OSError:
pass
# Get color image filenames
train_filenames, val_filenames, test_filenames = utils.get_color_file_names_by_bag(training_data_root,
which_bag=which_bag,
split_ratio=(0.5, 0.5))
training_folder_list, val_folder_list = utils.get_parent_folder_names(training_data_root, which_bag=which_bag)
# Build training and validation dataset
train_dataset = dataset.SfMDataset(image_file_names=train_filenames,
folder_list=training_folder_list + val_folder_list,
adjacent_range=adjacent_range, to_augment=True, transform=training_transforms,
downsampling=downsampling,
net_depth=teacher_depth, inlier_percentage=inlier_percentage,
use_store_data=load_intermediate_data,
store_data_root=precompute_root,
use_view_indexes_per_point=use_view_indexes_per_point, visualize=visualize,
phase="train", is_hsv=is_hsv)
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=shuffle,
num_workers=num_workers)
# Load trained teacher model
depth_estimation_model_teacher = models.UNet(in_channels=3, out_channels=1, depth=teacher_depth, wf=filter_base,
padding=True, up_mode='upsample')
log_root.mkdir(parents=True)
writer = SummaryWriter(logdir=str(log_root))
print("Tensorboard visualization at {}".format(str(log_root)))
if selected_frame_index_list is None and not load_all_frames:
raise IOError
# Read all frame indexes
if load_all_frames:
selected_frame_index_list = utils.read_visible_view_indexes(
sequence_root)
# Get color image filenames
test_filenames = utils.get_filenames_from_frame_indexes(
sequence_root, selected_frame_index_list)
folder_list = utils.get_parent_folder_names(evaluation_data_root,
id_range=id_range)
if phase == "validation":
test_dataset = dataset.SfMDataset(
image_file_names=test_filenames,
folder_list=folder_list,
adjacent_range=adjacent_range,
transform=None,
downsampling=input_downsampling,
network_downsampling=network_downsampling,
inlier_percentage=inlier_percentage,
use_store_data=load_intermediate_data,
store_data_root=evaluation_data_root,
phase="validation",
is_hsv=is_hsv,
num_pre_workers=num_pre_workers,
if not log_root.exists():
log_root.mkdir()
log_root = log_root / "dense_descriptor_train_{}_{}_{}_{}".format(
current_date.month, current_date.day, current_date.hour,
current_date.minute)
writer = SummaryWriter(logdir=str(log_root))
print("Created tensorboard visualization at {}".format(str(log_root)))
if not Path(args.precompute_root).exists():
Path(args.precompute_root).mkdir(parents=True)
train_filenames = \
utils.get_color_file_names_by_bag(root=Path(args.data_root), id_list=args.training_patient_id)
sequence_path_list = utils.get_parent_folder_names(
Path(args.data_root), id_list=args.training_patient_id)
# Build training and validation dataset
train_dataset = dataset.DescriptorDataset(
image_file_names=train_filenames,
folder_list=sequence_path_list,
adjacent_range=args.adjacent_range,
image_downsampling=args.image_downsampling,
inlier_percentage=args.inlier_percentage,
network_downsampling=args.network_downsampling,
load_intermediate_data=args.load_intermediate_data,
intermediate_data_root=Path(args.precompute_root),
sampling_size=args.sampling_size,
heatmap_sigma=args.heatmap_sigma,
num_pre_workers=args.num_workers,
visible_interval=args.visibility_overlap,