search

jd730 / OICR-pytorch

Pytorch Implementation of Multiple Instance Detection Network with Online Instance Classifier Refinement
MIT License
75 stars 10 forks source link

A question about your code #24

Closed lzrrr33 closed 4 years ago

lzrrr33 commented 4 years ago

hi bor ,thank you for sharing your code! when i read your code,i have a question: on line 124 of the /OICR-pytorch/lib/model/oicr/oicr.py , according to the paper, "self.ic_prob" should denote the proposal score matrices, but in your code, it seem like denote the classification score matrices. I don’t understand it,hope for your reply

jd730 commented 4 years ago

Hi @lzrrr33, ic_prob is a classification score for each proposal in the paper. Please see Eq. (5), refinement loss.

lzrrr33 commented 4 years ago

thank you for your reply ! but the item in self.mdin_prob is much lower than in self.ic_prob . this leads to the cls_loss_weight of ic0 lower than ic1 and ic2, is it normal?

jd730 commented 4 years ago

Hi @lzrrr33, Sorry about the late reply. I actually do not remember the scale of probs but what I remember is the losses from refinement branches are much lower than MIDN branch.

lzrrr33 commented 4 years ago

thanks，sorry to bother you