def resnext(
branch: str,
dataset_name: str,
embeddings_name: str,
batch_size: int,
workers: int,
n_gpu: int,
checkpoint: str,
date: str,
state_dict: str,
tag: List[str],
) -> None:
"""Create image embeddings with the ResNext model."""
from vscvs.models import ResNextNormalized
click.echo(
"Triplet ResNext embeddings for {} dataset".format(dataset_name))
model = load_triplet_model_from_checkpoint(ResNextNormalized,
ResNextNormalized,
ResNextNormalized, state_dict,
checkpoint, date, *tag)
embedding_network_model = {
"anchor": model.anchor_embedding_network,
"positive": model.positive_embedding_network,
"negative": model.negative_embedding_network,
}[branch]
create_embeddings(
embedding_network_model.base,
dataset_name,
embeddings_name,
batch_size,
workers,
n_gpu,
)
def hog(
dataset_name: str,
embeddings_name: str,
batch_size: int,
workers: int,
n_gpu: int,
in_channels: int,
cell_size: int,
bins: int,
signed_gradients: bool,
) -> None:
"""Create image embeddings with the HOG model."""
from vscvs.models import HOG
click.echo("HOG embeddings for {} dataset".format(dataset_name))
model = HOG(in_channels, cell_size, bins, signed_gradients)
create_embeddings(model, dataset_name, embeddings_name, batch_size, workers, n_gpu)
def cnn(
dataset_name: str,
embeddings_name: str,
batch_size: int,
workers: int,
n_gpu: int,
checkpoint: str,
date: str,
state_dict: str,
tag: List[str],
) -> None:
"""Create image embeddings with the CNN model."""
from vscvs.models import CNN
click.echo("CNN embeddings for {} dataset".format(dataset_name))
model = load_classification_model_from_checkpoint(
CNN, state_dict, checkpoint, date, *tag
)
model = remove_last_layer(model)
create_embeddings(model, dataset_name, embeddings_name, batch_size, workers, n_gpu)
def resnext(
dataset_name: str,
embeddings_name: str,
batch_size: int,
workers: int,
n_gpu: int,
checkpoint,
date,
state_dict,
tag,
) -> None:
"""Create image embeddings with the ResNext model."""
from vscvs.models import ResNext
click.echo("ResNext embeddings for {} dataset".format(dataset_name))
model = load_classification_model_from_checkpoint(
ResNext, state_dict, checkpoint, date, *tag
)
create_embeddings(
model.base, dataset_name, embeddings_name, batch_size, workers, n_gpu
)
def resnext(
branch: int,
dataset_name: str,
embeddings_name: str,
batch_size: int,
workers: int,
n_gpu: int,
checkpoint: str,
date: str,
state_dict: str,
tag: List[str],
) -> None:
"""Create image embeddings with the ResNext model."""
from vscvs.models import ResNextNormalized
click.echo(
"Siamese ResNext embeddings for {} dataset".format(dataset_name))
model = load_siamese_model_from_checkpoint(ResNextNormalized,
ResNextNormalized, state_dict,
checkpoint, date, *tag)
embedding_model = model.embedding_network_1 if branch else model.embedding_network_0
create_embeddings(embedding_model.base, dataset_name, embeddings_name,
batch_size, workers, n_gpu)