bolv.model.backbone

class ConvBlock

Parameters

• input_channels: int
• output_channels: int
• kernel_size: int, default=3
• stride: int, default=1
• padding: int, default=1
• use_bias: bool, default=True
• use_batch_norm: bool, default=True
• use_max_pool: bool, default=True
• max_padding: int, default=0
• activation: callable, default=nn.Relu

Methods

forward(input, params=None)

Conduct the forward propagation procedure.

Parameters: .

• input(Tensor) - The data input.

• params(Tensor, default=None) - Patameters of this block which will be used in forward propagation.

---

class ResBlock

Parameters

• input_channels: int
• num_filters: int
• max_pool: bool
• max_padding: int
• normalization: bool, default=True
• use_bias: bool, default=False

Methods

forward(input, params=None)

Conduct the forward propagation procedure.

Parameters: .

• input(Tensor) - The data input.

• params(Tensor, default=None) - Patameters of this block which will be used in forward propagation.

---

class Conv

Parameters

• input_shape: List[int]
• ways: int
• num_stages: int, default=4
• num_filters: int, default=64
• kernel_size: int, default=3
• stride: int, default=1
• padding: int, default=1
• use_max_pool: bool, default=True
• max_padding: int, default=0
• use_batch_norm: bool, default=True
• use_head: bool, default=True
• activation: callable, default=nn.Relu

Variables

• ~Conv.block{i}(nn.Module) - the block composing the backbone network, i=1,2...num_stages
• ~Conv.head(nn.Linear, default=None) - one FC layer at the end of network
• ~Conv.num_stages(int) - the num of conv blocks
• ~Conv.output_shape(Tensor) - the shape of output tensor

Methods

forward(input)

Conduct the forward propagation procedure.

Parameters:

• input(Tensor) - The data input.

---

class Res12

Parameters

• input_shape: List[int]
• ways: int
• use_head: bool, default=True

Variables

• ~Conv.block{i}(nn.Module) - the block composing the backbone network, i=1,2,3,4
• ~Conv.head(nn.Linear, default=None) - one FC layer at the end of network
• ~Conv.output_shape(Tensor) - the shape of output tensor

Methods

forward(input)

Conduct the forward propagation procedure.

Parameters:

• input(Tensor) - The data input.