I trained the KiTS23 dataset using nnUNet, and the model with the best EMA pseudo Dice of 0.9035 has a "foreground_mean" Dice of only 0.45634453650219475 during inference. The inference results are very poor, with many errors. What could be the problem?
I trained the KiTS23 dataset using nnUNet, and the model with the best EMA pseudo Dice of 0.9035 has a "foreground_mean" Dice of only 0.45634453650219475 during inference. The inference results are very poor, with many errors. What could be the problem?