Model Types

UNet - insar_eventnet.architectures.unet

The UNet model is the reccomended model for masking non-timeseries data, it utilizes a UNet architecture modified to use strided convolutions for downsampling instead of pooling.

UNet3D - insar_eventnet.architectures.unet3d

The corresponding model to UNet for masking timeseries data. This model is currently experimental.

ResNet - insar_eventnet.architectures.resnet

Not reccomended due to worse masking performance in comparison with UNet.

EventNet - insar_eventnet.architectures.eventnet

This is the binary prediction model which infers the presence of an event given a generated mask.