gingl / arn Goto Github PK

Hi;

I have read your paper and am carious about which dataset the faster-rcnn pre-trained on?

I have noticed that the original mattnet pre-trained the object detector on mscoco dateset and excluded all the val&test image in Refcoco, Refcocog, Refcoco+. Do you directly use the object detector in mattnet for extracting object proposals in this weakly supervised setting?

Hi
I have read your paper, and very interest in your approach in your paper.
but when I try to reproduce your result, I found out that I can not, the missing value makes me can not run your code, could you please tell me what is the exactly value of the exp_id? (--exp_id ${EXP_ID} in training and evaluation procedure) and adding it to your readme file? without this value, I can not go through your code and reproduce your result.