search

Maartendrinhuyzen / ai4mi_project

MIT License
0 stars 0 forks source link

Improve Loss Fn #3

Closed Maartendrinhuyzen closed 4 weeks ago

richtervanemmerik commented 1 month ago

Combing the CrossEntropy loss with DICE loss. Lecture Segmentation part 1 slide 66-69

richtervanemmerik commented 1 month ago

I Implemented it, but its pretty hard. This one was the most simple. Although tried training it but unfortunately DSC did not improve. Should we just present it still?

lisamenzel99 commented 1 month ago

I Implemented it, but its pretty hard. This one was the most simple. Although tried training it but unfortunately DSC did not improve. Should we just present it still?

I guess we could still present it. It is something we tried, but unfortunately did not have an effect. I don't know if it might have an effect later still once we have the data correct (aka heart class of the ground truth in the right place)