search

jiwoon-ahn / irn

Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations, CVPR 2019 (Oral)
MIT License
519 stars 100 forks source link

About the L_fg^D loss #7

Closed dongzhang89 closed 4 years ago

dongzhang89 commented 4 years ago

Thank you for the very good work! I have a question about the L_fg^D loss, in section 4.3, why the difference of (i, j) in D denotes D(x_i)-D(x_j) rather than D(x_j)-D(x_i)? I'm very confused about this point, looking forward to your reply. Thank you very much

jiwoon-ahn commented 4 years ago

It's simple, x_i + D(x_i) = x_j + D(x_j) must hold. Then, x_j - x_i = D(x_i) - D(x_j) follows.

dongzhang89 commented 4 years ago

OK, thank you very much!