search

yz93 / LAVT-RIS

GNU General Public License v3.0
185 stars 14 forks source link

About Dice Loss #28

Open nini0919 opened 1 year ago

nini0919 commented 1 year ago

Hello, author! Thank you for contributing an influential paper to the field of res. I have a small question to ask you regarding your statement on April 13th, 2023: "Using the Dice loss instead of the cross-entropy loss can improve results. Will add code and release weights later when getting a chance." Could you please provide the calculation code for the dice loss? Thank you, good luck!

yz93 commented 1 year ago

Hello! Hopefully I get time to do this soon. Thanks for your interest in the work.

nini0919 commented 1 year ago

Thank you for your patient answer. We are looking forward to your release of the code related to Dice loss.

Besides, I found the dimension of the network output tensor is [batch_size, 2, 480, 480], and I think the second dimension represents the classes of the pixel, 0 is background and 1 is foreground. Therefore, I want to ask you what is the meaning of 'multi-class' for multi-class Dice loss. Do you mean that "multi-class" refers to foreground and background classes?Thank you, good luck!