MONAI Implementation of Slice-Shift-UNet (SSH-UNet) Check our official code!!!!

[cugwu]

The ssh-unet repository (https://github.com/cugwu/ssh-unet) builds on insights from video action recognition to propose Slice Shift UNet (SSH-UNet), a 2D-based model that encodes 3D features with the efficiency of 2D CNNs.
It achieves this by applying 2D convolutions (conv-1x3x3) across three orthogonal planes while using weight sharing
to integrate multi-view features. The neglected third dimension is reintroduced via feature map shifting along the slice axis.

SSH-UNet is a ri-adaptation of MONAI's DynUNet, with most of the code based on the official MONAI implementation.

With only 6.48M parameters SSH-UNet surpasses famous methods like SwinUNETR and UNETR, which have 86.32M and 79.43M parameters, respectively, on BTCV and AMOS segmentation datasets when used under the same training conditions.

[pure-rgb]

Thanks for sharing. You seem like working on medical image data. I'm kinda new, could you please answer some of my preliminary quries over here #8391 TIA.