RNAN icon indicating copy to clipboard operation
RNAN copied to clipboard

Published 20 hours ago • verified publisher icon yulunzhang

Very impressed work, here are the problems about experiments.

Open raywzy opened this issue 6 years ago • 2 comments

In Table 1, case 3 denotes that there exits NLB but without Mask Branch. According to my understanding, NLB could not be used alone. Could you guys give me some explanations? Thx :D

raywzy avatar Jul 10 '19 05:07 raywzy

Hi, Let's see Figure 3, NLB didn't change the feature size (e.g., width, height, channel) and can be inserted into network structures, including mask branch. In Figure 2, we incorporate non-local block (NLB) into the mask branch to obtain non-local mixed attention. After removing mask branch in case 3, we directly treat the information from NLB as a 'mask'.

yulunzhang avatar Jul 12 '19 07:07 yulunzhang

One more question, why the weight ans bias of matrix W are set to 0?

https://github.com/yulunzhang/RNAN/blob/6377486f560d56dd9859f26217b455257fb70023/CAR/code/model/common.py#L91

raywzy avatar Jul 20 '19 08:07 raywzy