Reproducing the perfomance

[gwang-kim]

Hi, thanks for the great work.
I tried reproducing your performance following the instruction. (Albunet_valid_1024)

After part 0 training, the best performance for 5-folds looks just 0.78 ( I checked experiment/almunet_valid/checkpoint/summary.csv), which is lower than I think (0.85),
could you let me know how to get 0.85 of dice score on validation set?