graph LR
ResNet["ResNet"]
WideResNet["WideResNet"]
Xception["Xception"]
ResNet_Variants["ResNet Variants"]
Backbone_Instantiation_Selection["Backbone Instantiation/Selection"]
ResNet_Variants -- "instantiates" --> ResNet
Backbone_Instantiation_Selection -- "depends on" --> ResNet
Backbone_Instantiation_Selection -- "depends on" --> WideResNet
Backbone_Instantiation_Selection -- "depends on" --> Xception
The Feature Extraction Backbones subsystem is primarily defined by the encoding.models.backbone package. This package encapsulates various deep learning architectures designed for extracting hierarchical features from input data.
Implements the foundational ResNet architecture, including its core building blocks (_make_layer) and overall network structure. It serves as a base class for creating deep residual networks.
Related Classes/Methods:
Implements the Wide ResNet architecture, which modifies the standard ResNet by increasing the width (number of filters) of layers to enhance performance while maintaining a relatively shallow depth.
Related Classes/Methods:
Implements the Xception architecture, which efficiently extracts features using depthwise separable convolutions. It includes unique building blocks like SeparableConv2d and Block.
Related Classes/Methods:
Provides pre-configured and optimized instances of the ResNet architecture. These functions simplify the instantiation of common ResNet models tailored for specific performance or accuracy requirements.
Related Classes/Methods:
A higher-level module responsible for instantiating and retrieving a chosen backbone (ResNet, WideResNet, or Xception) based on configuration. It acts as an interface for consumers to obtain a specific feature extraction backbone.
Related Classes/Methods: