Skip to content

Semantic_Segmentation_Models.md

Latest commit

 

History

History
111 lines (65 loc) · 6.34 KB

Semantic_Segmentation_Models.md

File metadata and controls

111 lines (65 loc) · 6.34 KB
graph LR
    BaseNet["BaseNet"]
    ATTEN["ATTEN"]
    EncNet["EncNet"]
    DeepLabV3["DeepLabV3"]
    PSP["PSP"]
    FCN["FCN"]
    UperNet["UperNet"]
    Segmentation_Heads["Segmentation Heads"]
    ATTEN -- "inherits from" --> BaseNet
    EncNet -- "inherits from" --> BaseNet
    DeepLabV3 -- "inherits from" --> BaseNet
    PSP -- "inherits from" --> BaseNet
    FCN -- "inherits from" --> BaseNet
    UperNet -- "inherits from" --> BaseNet
    ATTEN -- "composes" --> Segmentation_Heads
    EncNet -- "composes" --> Segmentation_Heads
    DeepLabV3 -- "composes" --> Segmentation_Heads
    PSP -- "composes" --> Segmentation_Heads
    FCN -- "composes" --> Segmentation_Heads
    UperNet -- "composes" --> Segmentation_Heads
Loading

CodeBoardingDemoContact

Details

The Semantic Segmentation Models subsystem (encoding.models.sseg) is responsible for implementing various advanced semantic segmentation architectures that build upon extracted features for pixel-level classification.

BaseNet

Serves as the foundational abstract class for all semantic segmentation models. It encapsulates common functionalities such as initializing the backbone network and orchestrating the generic inference pipeline, including image transformations and delegating to module_inference.

Related Classes/Methods:

ATTEN

Implements the ATTEN semantic segmentation architecture, inheriting from BaseNet. It is responsible for composing its unique ATTENHead and defining any architecture-specific layers or connections beyond the backbone.

Related Classes/Methods:

EncNet

Implements the EncNet semantic segmentation architecture, inheriting from BaseNet. It is responsible for composing its unique EncHead and defining any architecture-specific layers or connections beyond the backbone.

Related Classes/Methods:

DeepLabV3

Implements the DeepLabV3 semantic segmentation architecture, inheriting from BaseNet. It is responsible for composing its unique DeepLabV3Head and integrating specific modules like Atrous Spatial Pyramid Pooling (ASPP).

Related Classes/Methods:

PSP

Implements the PSP (Pyramid Scene Parsing Network) semantic segmentation architecture, inheriting from BaseNet. It reuses the FCNHead for its final predictions.

Related Classes/Methods:

FCN

Implements the FCN (Fully Convolutional Network) semantic segmentation architecture, inheriting from BaseNet. It is responsible for composing its specific FCNHead for predictions.

Related Classes/Methods:

UperNet

Implements the UperNet (Unified Perceptual Parsing Network) semantic segmentation architecture, inheriting from BaseNet. It is responsible for composing its specific UperNetHead for predictions.

Related Classes/Methods:

Segmentation Heads

This component represents a collection of specialized modules (e.g., ATTENHead, EncHead, DeepLabV3Head, FCFPNHead, PSPHead, FCNHead, UperNetHead) that process features extracted by the backbone network to produce the final pixel-level predictions specific to their respective segmentation architectures.

Related Classes/Methods:

FAQ