not converge with custom dataset

[Chq2018515]

thanks for your great work! I was using pretrained model R50-ViT-B_16 and trained it on some opensource dataset for lung segmentation. During training, the total loss stayed above 0.7 and ce loss stayed at 0.3. And in testing phase, i got results like "mean_dice : 0.354955 mean_hd95 : 77.297413 mean_precision : 0.373490". I wonder what i can do with it. Any help would be appreciated : )