This is the code of the semester project for the course "Machine learning for 3D Geometry" of the Technical University of Munich.
The main branch contains the base nnU-Net framework. We made 5 modifications and saved each of them in their respective branch.
Branch | Description |
---|---|
Main | Base nnU-Net framework without any modification. |
attention_nnunet | Transformed the Vanilla U-Net into an Attention U-Net while mantaining the dynamic network creation. |
stacked_dilated_convolutions | Added in each level of the Vanilla U-Net stacked dilated convolutions while mantaining the dynamic network creation. |
decreasingly_dilated_convolutions | Added in each encoder level of the Vanilla U-Net decreasingly dilated convolutions while mantaining the dynamic network creation. |
nnunet_3plus | Transformed the Vanilla U-Net into an U-Net 3+ while mantaining the dynamic network creation. |
model_quantization | Techniques to reduce model footprint. |
Please refer to the original nnU-Net repository for the installation guide.