insar_eventnet.architectures package

Submodules

insar_eventnet.architectures.eventnet module

Summary

Basic convolutional model, for now, for classifying the images.

Notes

Created By: Andrew Player

insar_eventnet.architectures.eventnet.conv2d_block(input_tensor: Tensor, num_filters: int, kernel_size: int = 3, strides: int = 1) → Tensor

2D-Convolution Block for encoding / generating feature maps.

insar_eventnet.architectures.eventnet.create_eventnet(model_name: str = 'model', tile_size: int = 512, num_filters: int = 32, label_count: int = 1, learning_rate: float = 0.005) → Model

Creates a basic convolutional network

insar_eventnet.architectures.resnet module

Summary

Contains a network for sar classification

References

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7041455/

Notes

Created By: Andrew Player

insar_eventnet.architectures.resnet.create_resnet(model_name: str = 'model', tile_size: int = 512, num_filters: int = 4, learning_rate: float = 0.0001) → Model

Creates a model for unwrapping 2D wrapped phase images.

References

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7041455/

insar_eventnet.architectures.resnet.full_block(input_tensor: Tensor, num_filters: int) → Tensor

Sequential Block with a 2D-Convolution into MaxPooling, followed by 4 Residual Convolution Blocks.

insar_eventnet.architectures.resnet.res4_block(input_tensor: Tensor, num_filters: int) → Tensor

Sequential block of 4 Residual Convolution Blocks

insar_eventnet.architectures.resnet.res_block(input_tensor: Tensor, num_filters: int) → Tensor

2D-Convolution Block with a connecting convolution for short-term memory.

insar_eventnet.architectures.resnet_classifier module

Summary

Contains a network for sar classification

References

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7041455/

Notes

Created By: Andrew Player

insar_eventnet.architectures.resnet_classifier.conv_block(input_tensor: Tensor, num_filters: int) → Tensor

2D-Convolution Block with a connecting convolution for short-term memory.

insar_eventnet.architectures.resnet_classifier.create_resnetclassifier(model_name: str = 'model', tile_size: int = 512, num_filters: int = 4, learning_rate: float = 0.0001) → Model

Creates a model for unwrapping 2D wrapped phase images.

References

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7041455/

insar_eventnet.architectures.resnet_classifier.identity_block(input_tensor: Tensor, num_filters: int)

insar_eventnet.architectures.unet module

Summary

Contains a network for sar classification

Notes

Created By: Andrew Player

insar_eventnet.architectures.unet.conv2d_block(input_tensor: Tensor, num_filters: int, kernel_size: int = 3, strides: int = 1) → Tensor

UNET style 2D-Convolution Block for encoding / generating feature maps.

insar_eventnet.architectures.unet.create_unet(model_name: str = 'model', tile_size: int = 512, num_filters: int = 64, learning_rate: float = 0.0001) → Model

Creates a U-Net style model

insar_eventnet.architectures.unet.transpose_block(input_tensor: Tensor, concat_tensor: Tensor, num_filters: int) → Tensor

Learned Upscaling for decoding

insar_eventnet.architectures.unet3d module

Summary

Contains a network for sar classification

Notes

Created By: Andrew Player

insar_eventnet.architectures.unet3d.conv3d_block(input_tensor: Tensor, num_filters: int, kernel_size: int = 3, strides: int = 1) → Tensor

UNET style 3D-Convolution Block for encoding / generating feature maps.

insar_eventnet.architectures.unet3d.create_unet3d(model_name: str = 'model', tile_size: int = 512, temporal_steps: int = 16, num_filters: int = 16, learning_rate: float = 0.0001) → Model

Creates a 3D U-Net style model

insar_eventnet.architectures.unet3d.transpose_block(input_tensor: Tensor, concat_tensor: Tensor, num_filters: int) → Tensor

Learned Upscaling for decoding

Module contents