regnet_params

Typed parameters of the regnet family; importable without Keras.

RegNetBlockParam()

RegNet block parameters

model_config = ConfigDict(frozen=True, extra='forbid')

filters: int = Field(..., description='# filters')

Number of filters

depth: int = Field(default=1, description='Layer depth')

Layer depth

group_width: int = Field(default=1, description='Group width. Must be divisible by in/out filters')

Group width

kernel_size: int | tuple[int, int] = Field(default=3, description='Kernel size')

Kernel size

strides: int | tuple[int, int] = Field(default=1, description='Stride size')

Stride size

se_ratio: float = Field(default=8, description='Squeeze Excite ratio')

Squeeze Excite ratio

droprate: float = Field(default=0, description='Drop rate')

Drop rate

activation: str = Field(default='relu6', description='Activation function')

Activation function

RegNetParams()

RegNet parameters

model_config = ConfigDict(frozen=True, extra='forbid')

family: Literal['regnet'] = 'regnet'

Model family

blocks: list[RegNetBlockParam] = Field(default_factory=list, description='RegNet blocks')

RegNet blocks

input_filters: int = Field(default=0, description='Input filters')

Input filters

input_strides: int | tuple[int, int] = Field(default=2, description='Input stride')

Input stride

input_activation: str = Field(default='relu6', description='Input activation')

Input activation

output_filters: int = Field(default=0, description='Output filters')

Output filters

block_style: Literal['y', 'z'] = Field(default='y', description='Block style')

Block style

include_top: bool = Field(default=True, description='Include top')

Include top

num_classes: int | None = Field(default=None, gt=0, description='Classes of the output layer')

Classes of the output layer

output_activation: str | None = Field(default=None, description='Output activation')

Output activation

dropout: float = Field(default=0.2, description='Dropout rate')

Dropout rate