Hi, thanks for sharing your research with us. I implemented kiUNet-3D for my application, had to modify the architecture so that I can train the model. But the training is exceptionally slow. For an input of size 140 x 140 x 140 and dataset of 40 volumes, it takes more than 30 minutes per epoch. I have played with many other 3d architectures but I have observed KiUNet's training is quite slow. Could you suggest any way to speed-up the training, I have already reduced the model.