search

zhang-pengyu / ADRNet

code and results for 'Learning Adaptive Attribute-Driven Representation for Real-Time RGB-T Tracking'
45 stars 4 forks source link

关于多步训练 #11

Closed caiyujue closed 2 years ago

caiyujue commented 2 years ago

我想请问下,在论文里的分步训练的第二步训练AENet时,论文里说要冻结主干.ADRB和FC,这里的FC层,是FC4.FC5.FC6吗,包不包括FC6啊,还有就是怎么最小化公式6的那个权重Ma,这边没怎么看明白,希望作者大大能看到,回复一下

zhang-pengyu commented 2 years ago

你好，不需要冻结FC6，另外第二步也是优化第一步里（eq.8）来得到权重Ma