Hyper U-Net ResNet34¶

class src.pipelines.tensorflow_v2.models.hyper_unet_resnet.ResBlock(channels, kernel_size, activation, initializer, stride=1, decode=False)[source]¶

A Residual Block. This is similar to the class defined in the unet_resnet.py script, but has the addition of the kernel_size in the instantiator to allow the option to include this parameter in the search space.

call(x)[source]¶

Applies a ResBlock to the an input

Parameters: x (Tensor) – the input to apply the ResBlock to
Return type: Tensor
Returns: the output of the ResBlock

class src.pipelines.tensorflow_v2.models.hyper_unet_resnet.HyperUnetResnet(input_shape, output_channels)[source]¶

A Hyper U-Net ResNet34 model which can be tuned by kerastuner

build(hp)[source]¶

A function which creates a kerastuner hyper model with a defined search space. This function signature matches the requirements of the hyperparameter tuning algorithms in kerastuner.

Parameters: hp (HyperParameters) – a HyperParameters instance
Return type: HyperModel
Returns: a HyperUnetResnet instance