xianlin7 / SAMUS

MIT License
152 stars 13 forks source link

用US30K数据集训练时,在113个epoch trainloss和val loss都突然变得很高,后面dice分数一直很低,是怎么回事 #19

Open ncp-nice opened 6 months ago

ncp-nice commented 6 months ago

image image

Ianyliu commented 4 months ago

@ncp-nice 請問您是自行整理和匯合得到US30K數據嗎?我好像找不到公開的整理好的數據集

Baldwinzc commented 1 month ago

Hello, I have also encountered this problem. Have you solved it?