adjacent_range=adjacent_range,
transform=None,
downsampling=input_downsampling,
network_downsampling=network_downsampling,
inlier_percentage=inlier_percentage,
use_store_data=True,
store_data_root=training_data_root,
phase="validation", is_hsv=is_hsv,
num_pre_workers=num_workers, visible_interval=30, rgb_mode="rgb")
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True,
num_workers=num_workers)
validation_loader = torch.utils.data.DataLoader(dataset=validation_dataset, batch_size=batch_size, shuffle=False,
num_workers=batch_size)
depth_estimation_model_student = models.FCDenseNet57(n_classes=1)
# Initialize the depth estimation network with Kaiming He initialization
depth_estimation_model_student = utils.init_net(depth_estimation_model_student, type="kaiming", mode="fan_in",
activation_mode="relu",
distribution="normal")
# Multi-GPU running
depth_estimation_model_student = torch.nn.DataParallel(depth_estimation_model_student)
# Summary network architecture
if display_architecture:
torchsummary.summary(depth_estimation_model_student, input_size=(3, height, width))
# Optimizer
optimizer = torch.optim.SGD(depth_estimation_model_student.parameters(), lr=max_lr, momentum=0.9)
lr_scheduler = scheduler.CyclicLR(optimizer, base_lr=min_lr, max_lr=max_lr, step_size=num_iter)
# Custom layers
depth_scaling_layer = models.DepthScalingLayer(epsilon=depth_scaling_epsilon)
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,
visible_interval=visibility_overlap,
rgb_mode="rgb")
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=0)
depth_estimation_model = models.FCDenseNet57(n_classes=1)
# Initialize the depth estimation network with Kaiming He initialization
utils.init_net(depth_estimation_model,
type="kaiming",
mode="fan_in",
activation_mode="relu",
distribution="normal")
# Multi-GPU running
depth_estimation_model = torch.nn.DataParallel(depth_estimation_model)
# Summary network architecture
if display_architecture:
torchsummary.summary(depth_estimation_model,
input_size=(3, height, width))
# Load trained model
if trained_model_path.exists():