graph LR
UNet["UNet"]
HighRes3DNet["HighRes3DNet"]
DenseVNet["DenseVNet"]
ConvLayer["ConvLayer"]
DeconvLayer["DeconvLayer"]
BaseLayer["BaseLayer"]
UNet -- "utilizes" --> ConvLayer
UNet -- "utilizes" --> DeconvLayer
HighRes3DNet -- "utilizes" --> ConvLayer
DenseVNet -- "utilizes" --> ConvLayer
ConvLayer -- "inherits from" --> BaseLayer
DeconvLayer -- "inherits from" --> BaseLayer
The Network Architectures (Model Zoo) subsystem in NiftyNet is responsible for providing a collection of pre-defined deep learning models and fundamental building blocks for constructing various network architectures, primarily for medical imaging tasks.
Implements the U-Net convolutional network architecture, commonly used for biomedical image segmentation. It composes UNetBlocks, which in turn utilize fundamental layers.
Related Classes/Methods:
Implements the HighRes3DNet architecture, designed for high-resolution 3D medical image analysis. It composes HighResBlocks, which utilize fundamental layers.
Related Classes/Methods:
Implements the DenseVNet architecture, characterized by dense connectivity patterns for efficient feature reuse. It composes DenseFeatureStackBlocks, which utilize fundamental layers.
Related Classes/Methods:
A fundamental building block for N-dimensional convolution operations, serving as a core component for feature extraction in most network architectures.
Related Classes/Methods:
A fundamental building block for N-dimensional deconvolution (transposed convolution) operations, typically used for upsampling in network architectures like U-Net.
Related Classes/Methods:
Provides a common interface and shared functionalities for all custom layers within the framework, ensuring consistency and extensibility across different network building blocks.
Related Classes/Methods: