Vision_Transformer¶

VISION_TRANSFORMER¶

Segformer¶

        classDiagram
    class Segformer {
        +int num_spatial_dims
        +bool register
        +forward() Tensor
        +validate_input_shape() tuple
    }
    class BaseModel {
    }
    <<abstract>> BaseModel
    BaseModel <|-- Segformer : herits
    class ModelABC {
        +bool register
        +int in_channels
        +int out_channels
        +tuple input_shape
        +check_required_attributes() None
    }
    <<abstract>> ModelABC
    ModelABC <|-- Segformer : herits
    class Module
    <<abstract>> Module
    Module <|-- Segformer : herits

Segformer

Segformer architecture with extra upsampling in the decoder to match the input image size.

SwinUNetR¶

        classDiagram
    class SwinUNetR {
        +bool onnx_supported
        +tuple supported_num_spatial_dims
        +bool features_last
        +ModelType model_type
        +int num_spatial_dims
        +bool register
        +forward() Tensor
        +validate_input_shape() tuple
    }
    class ModelABC {
        +bool register
        +int in_channels
        +int out_channels
        +tuple input_shape
        +check_required_attributes() None
    }
    <<abstract>> ModelABC
    ModelABC <|-- SwinUNetR : herits
    class Module
    <<abstract>> Module
    Module <|-- SwinUNetR : herits

SwinUNetR

Wrapper around the SwinUNETR from MONAI.

UNetRPP¶

        classDiagram
    class UNetRPP {
        +bool features_last
        +bool register
        +forward() torch.Tensor | list[torch.Tensor]
        +proj_feat() Tensor
        +validate_input_shape() tuple
    }
    class BaseModel {
    }
    <<abstract>> BaseModel
    BaseModel <|-- UNetRPP : herits
    class ModelABC {
        +bool register
        +int in_channels
        +int out_channels
        +tuple input_shape
        +check_required_attributes() None
    }
    <<abstract>> ModelABC
    ModelABC <|-- UNetRPP : herits
    class Module
    <<abstract>> Module
    Module <|-- UNetRPP : herits

UNetRPP

UNetR++ based on: "Shaker et al., UNETR++: Delving into Efficient and Accurate 3D Medical Image Segmentation".

ViTClassifier¶

        classDiagram
    class ViTClassifier {
        +bool onnx_supported
        +tuple supported_num_spatial_dims
        +bool features_last
        +ModelType model_type
        +int num_spatial_dims
        +bool register
        +forward() Tensor
    }
    class BaseModel {
    }
    <<abstract>> BaseModel
    BaseModel <|-- ViTClassifier : herits
    class ModelABC {
        +bool register
        +int in_channels
        +int out_channels
        +tuple input_shape
        +check_required_attributes() None
    }
    <<abstract>> ModelABC
    ModelABC <|-- ViTClassifier : herits
    class Module
    <<abstract>> Module
    Module <|-- ViTClassifier : herits

ViTClassifier

Vision Transformer (ViT) classifier model outputing class probabilities per input sample.

VitEncoder¶

        classDiagram
    class VitEncoder {
        +bool onnx_supported
        +tuple supported_num_spatial_dims
        +bool features_last
        +ModelType model_type
        +int num_spatial_dims
        +bool register
        +forward() Tensor
    }
    class BaseModel {
    }
    <<abstract>> BaseModel
    BaseModel <|-- VitEncoder : herits
    class ModelABC {
        +bool register
        +int in_channels
        +int out_channels
        +tuple input_shape
        +check_required_attributes() None
    }
    <<abstract>> ModelABC
    ModelABC <|-- VitEncoder : herits
    class Module
    <<abstract>> Module
    Module <|-- VitEncoder : herits

VitEncoder

ViT vision encoder for multimodal LLMs.