Closed rginpan closed 2 years ago
@rginjapan communicate only in English please, so that every can understand questions and reply accordingly. Thanks
@arkinrc Thanks for being interested in my question, lol. I am right now implementing CE to my framework and using this repo. as reference. If I understand correctly, Channel Exchange will make one feature vector becomes more and more important but the other one becomes more and more meaningless, why in the end of MLF-VO, it still fuse A and B together instead of using the more important feature vector. As shown in Fig2(a), after the last CE, there are still using information from both streams.
你好,想问一个关于Channel exchange 的问题,A和B 做 CE后, 假设A变得越来越important,B越来越不important,为什么最后还是fusionA和B呢, 而不是只用A??