ConvNeXt Networks#
Classes#
- class ConvNeXt(name, in_shape, out_shape, *, max_drop_path=0, layer_scale=1e-06, dims=None, depths=None)[source]#
Bases:
NetworkConvNeXt network using Network.
Original paper implementation: arXiv:2201.03545.
- Attributes:
- layer_num
int,Optional Number of layers to use, if None use all layers
- group
int Which group is the active group if a layer has the group attribute
- name
str Name of the network, used for saving
- version
str NetLoader version string
- checkpoints
list[TensorListLike] Outputs from each checkpoint with each Tensor having shape (N,…) and type float, where N is the batch size
- config
dict[str,Any] Network configuration
- kl_loss
Tensor KL divergence loss on the latent space, of shape (1) and type float, if using a sample layer
- net
TypedModuleList Network construction
- shapes
Shapes Layer output shapes
- check_shapes
Shapes Checkpoint output shapes
- layer_num
- Parameters:
- name
str Name of the network configuration file
- in_shape
list[int] |tuple[int, …] Shape of the input tensor, excluding batch size
- out_shape
list[int] shape of the output tensor, excluding batch size
- max_drop_path
float,Optional Maximum drop path fraction, default = 0
- layer_scale
float,Optional Default value for learnable parameters to scale the convolutional filters in a ConvNeXt block, default = 1e-6
- dims
list[int] |NoneType,Optional Number of convolutional filters in each downscaled section of the network, default = [96, 192, 384, 768]
- depths
list[int] |NoneType,Optional Number of ConvNeXt blocks in each downscaled section of the network, default = [3, 3, 9, 3]
- name
- class ConvNeXtTiny(in_shape, out_shape, *, max_drop_path=0, layer_scale=1e-06)[source]#
Bases:
ConvNeXtTiny version of ConvNeXt with the number of convolutional filters in each downscaled section following [96, 192, 384, 768] with the number of ConvNeXt blocks in each downscaled section following [3, 3, 9, 3].
Original paper implementation: arXiv:2201.03545.
- Attributes:
- layer_num
int,Optional Number of layers to use, if None use all layers
- group
int Which group is the active group if a layer has the group attribute
- name
str Name of the network, used for saving
- version
str NetLoader version string
- checkpoints
list[TensorListLike] Outputs from each checkpoint with each Tensor having shape (N,…) and type float, where N is the batch size
- config
dict[str,Any] Network configuration
- kl_loss
Tensor KL divergence loss on the latent space, of shape (1) and type float, if using a sample layer
- net
TypedModuleList Network construction
- shapes
Shapes Layer output shapes
- check_shapes
Shapes Checkpoint output shapes
- layer_num
- Parameters:
- in_shape
list[int] |list[list[int]] |tuple[int, …] Shape of the input tensor(s), excluding batch size
- out_shape
list[int] shape of the output tensor, excluding batch size
- max_drop_path
float,Optional Maximum drop path fraction, default = 0
- layer_scale
float,Optional Default value for learnable parameters to scale the convolutional filters in a ConvNeXt block, default = 1e-6
- in_shape
- class ConvNeXtSmall(in_shape, out_shape, *, max_drop_path=0, layer_scale=1e-06)[source]#
Bases:
ConvNeXtSmall version of ConvNeXt with the number of convolutional filters in each downscaled section following [96, 192, 384, 768] with the number of ConvNeXt blocks in each downscaled section following [3, 3, 27, 3].
Original paper implementation: arXiv:2201.03545.
- Attributes:
- layer_num
int,Optional Number of layers to use, if None use all layers
- group
int Which group is the active group if a layer has the group attribute
- name
str Name of the network, used for saving
- version
str NetLoader version string
- checkpoints
list[TensorListLike] Outputs from each checkpoint with each Tensor having shape (N,…) and type float, where N is the batch size
- config
dict[str,Any] Network configuration
- kl_loss
Tensor KL divergence loss on the latent space, of shape (1) and type float, if using a sample layer
- net
TypedModuleList Network construction
- shapes
Shapes Layer output shapes
- check_shapes
Shapes Checkpoint output shapes
- layer_num
- Parameters:
- in_shape
list[int] |list[list[int]] |tuple[int, …] Shape of the input tensor(s), excluding batch size
- out_shape
list[int] shape of the output tensor, excluding batch size
- max_drop_path
float,Optional Maximum drop path fraction, default = 0
- layer_scale
float,Optional Default value for learnable parameters to scale the convolutional filters in a ConvNeXt block, default = 1e-6
- in_shape
- class ConvNeXtBase(in_shape, out_shape, *, max_drop_path=0, layer_scale=1e-06)[source]#
Bases:
ConvNeXtBase version of ConvNeXt with the number of convolutional filters in each downscaled section following [128, 256, 512, 1024] with the number of ConvNeXt blocks in each downscaled section following [3, 3, 27, 3].
Original paper implementation: arXiv:2201.03545.
- Attributes:
- layer_num
int,Optional Number of layers to use, if None use all layers
- group
int Which group is the active group if a layer has the group attribute
- name
str Name of the network, used for saving
- version
str NetLoader version string
- checkpoints
list[TensorListLike] Outputs from each checkpoint with each Tensor having shape (N,…) and type float, where N is the batch size
- config
dict[str,Any] Network configuration
- kl_loss
Tensor KL divergence loss on the latent space, of shape (1) and type float, if using a sample layer
- net
TypedModuleList Network construction
- shapes
Shapes Layer output shapes
- check_shapes
Shapes Checkpoint output shapes
- layer_num
- Parameters:
- in_shape
list[int] |list[list[int]] |tuple[int, …] Shape of the input tensor(s), excluding batch size
- out_shape
list[int] shape of the output tensor, excluding batch size
- max_drop_path
float,Optional Maximum drop path fraction, default = 0
- layer_scale
float,Optional Default value for learnable parameters to scale the convolutional filters in a ConvNeXt block, default = 1e-6
- in_shape
- class ConvNeXtLarge(in_shape, out_shape, *, max_drop_path=0, layer_scale=1e-06)[source]#
Bases:
ConvNeXtLarge version of ConvNeXt with the number of convolutional filters in each downscaled section following [192, 384, 768, 1536] with the number of ConvNeXt blocks in each downscaled section following [3, 3, 27, 3].
Original paper implementation: arXiv:2201.03545.
- Attributes:
- layer_num
int,Optional Number of layers to use, if None use all layers
- group
int Which group is the active group if a layer has the group attribute
- name
str Name of the network, used for saving
- version
str NetLoader version string
- checkpoints
list[TensorListLike] Outputs from each checkpoint with each Tensor having shape (N,…) and type float, where N is the batch size
- config
dict[str,Any] Network configuration
- kl_loss
Tensor KL divergence loss on the latent space, of shape (1) and type float, if using a sample layer
- net
TypedModuleList Network construction
- shapes
Shapes Layer output shapes
- check_shapes
Shapes Checkpoint output shapes
- layer_num
- Parameters:
- in_shape
list[int] |list[list[int]] |tuple[int, …] Shape of the input tensor(s), excluding batch size
- out_shape
list[int] shape of the output tensor, excluding batch size
- max_drop_path
float,Optional Maximum drop path fraction, default = 0
- layer_scale
float,Optional Default value for learnable parameters to scale the convolutional filters in a ConvNeXt block, default = 1e-6
- in_shape
- class ConvNeXtXLarge(in_shape, out_shape, *, max_drop_path=0, layer_scale=1e-06)[source]#
Bases:
ConvNeXtExtra large version of ConvNeXt with the number of convolutional filters in each downscaled section following [256, 512, 1024, 2048] with the number of ConvNeXt blocks in each downscaled section following [3, 3, 27, 3].
Original paper implementation: arXiv:2201.03545.
- Attributes:
- layer_num
int,Optional Number of layers to use, if None use all layers
- group
int Which group is the active group if a layer has the group attribute
- name
str Name of the network, used for saving
- version
str NetLoader version string
- checkpoints
list[TensorListLike] Outputs from each checkpoint with each Tensor having shape (N,…) and type float, where N is the batch size
- config
dict[str,Any] Network configuration
- kl_loss
Tensor KL divergence loss on the latent space, of shape (1) and type float, if using a sample layer
- net
TypedModuleList Network construction
- shapes
Shapes Layer output shapes
- check_shapes
Shapes Checkpoint output shapes
- layer_num
- Parameters:
- in_shape
list[int] |list[list[int]] |tuple[int, …] Shape of the input tensor(s), excluding batch size
- out_shape
list[int] shape of the output tensor, excluding batch size
- max_drop_path
float,Optional Maximum drop path fraction, default = 0
- layer_scale
float,Optional Default value for learnable parameters to scale the convolutional filters in a ConvNeXt block, default = 1e-6
- in_shape