MonAI Unet -> linear output

[System-Lord22]

I've been using a MinAI UNet to examine an image and then an AdaptiveAvgPool3d to put it into a number of channels to go into a fully connected layer and then give a linear output (a "score").
It seems to work alright but someone said it was more traditional to just capture the middle layer of the UNet as the input to the FC layer.

Any opinions? And how would I easily chop the UNet in half like that?

Thanks

[aylward]

The appropriateness of this depends on your task.

To get at the middle/bottom layer of a UNet, here is one approach: Project-MONAI/tutorials#949

[System-Lord22]

Can I get access to the UNet forward function from somewhere so I can subclass and override it?