efficientnet

EfficientNetV2

Overview

EfficientNetV2 is an improvement to EfficientNet that incorporates additional optimizations to reduce both computation and memory. In particular, the architecture leverages both fused and non-fused MBConv blocks, non-uniform layer scaling, and training-aware NAS.

For more info, refer to the original paper EfficientNetV2: Smaller Models and Faster Training.

Parameters are in helia_edge.models.efficientnet_params.

Functions

Name	Description
`build`	EfficientNetV2 model from `EfficientNetParams`
`efficientnetv2_layer`	EfficientNetV2 layer

Additions

The EfficientNetV2 architecture has been modified to allow the following:

Enable 1D and 2D variants.

Usage

from helia_edge.layers import MBConvParams
from helia_edge.models import EfficientNetParams, ModelSpec, build

params = EfficientNetParams(
    input_filters=24,
    input_kernel_size=(1, 7),
    input_strides=(1, 2),
    blocks=[
        MBConvParams(filters=32, depth=2, kernel_size=(1, 7), strides=(1, 2), ex_ratio=1, se_ratio=2),
        MBConvParams(filters=48, depth=2, kernel_size=(1, 7), strides=(1, 2), ex_ratio=1, se_ratio=2),
        MBConvParams(filters=64, depth=2, kernel_size=(1, 7), strides=(1, 2), ex_ratio=1, se_ratio=2),
        MBConvParams(filters=72, depth=1, kernel_size=(1, 7), strides=(1, 2), ex_ratio=1, se_ratio=2),
    ],
    output_filters=0,
    include_top=True,
    num_classes=5,
    dropout=0.2,
    drop_connect_rate=0.2,
)
model = build(ModelSpec(params=params, input_shape=(1, 800, 1)))

Machine-readable model

efficientnet_corefunctionEfficientNet core
efficientnetv2_layerfunctionCreate EfficientNet V2 TF functional model
buildfunctionBuild a EfficientNetV2 model.

efficientnet_core(blocks: list[MBConvParams], drop_connect_rate: float = 0) -> keras.Layer

EfficientNet core

Parameters

Parameters of efficientnet_core
Name	Type	Default	Description
`blocks`	`list[MBConvParam]`	Required	MBConv params
`drop_connect_rate`	`float`	0	Drop connect rate. Defaults to 0.

Returns

Returns of efficientnet_core
Type	Description
`keras.Layer`	keras.Layer: Core

efficientnetv2_layer(x: keras.KerasTensor, params: EfficientNetParams) -> keras.KerasTensor

Create EfficientNet V2 TF functional model

Parameters

Parameters of efficientnetv2_layer
Name	Type	Default	Description
`x`	`keras.KerasTensor`	Required	Input tensor
`params`	`EfficientNetParams`	Required	Model parameters.

Returns

Returns of efficientnetv2_layer
Type	Description
`keras.KerasTensor`	keras.KerasTensor: Output tensor

build(
    params: EfficientNetParams,
    input_shape: tuple[int | None, ...],
    *,
    batch_size: int | None = None,
    name: str | None = None,
) -> keras.Model

Build a EfficientNetV2 model.

Parameters

Parameters of build
Name	Type	Default	Description
`params`	`EfficientNetParams`	Required	Model parameters.
`input_shape`	`tuple[int \| None, ...]`	Required	Input shape without the batch axis; None for a variable axis.
`batch_size`	`int \| None`	None	Static batch size; None for a dynamic batch.
`name`	`str \| None`	None	Model name; the family when None.

Returns

Returns of build
Type	Description
`keras.Model`	keras.Model: The model, named ``efficientnet`` unless ``name`` is given.