This is a dense-unet segmentation architecture based on pretained dense net.
We created three dense-unet architecture based on DenseNet121, DenseNet161 and DenseNet201, the details of the proposed dense-unet is listed in the following:
We test U-Net, Res-Unet, DenseUnet121, DenseUnet161, DenseUnet201 in this project, training loss and validation dice similarity coeficient are shown below:
python train.py --save ./work/super/test
python test.py --resume $resume_path$