z-hXu/ReCU

Issues

• evaluation error

  #11 opened 2 years ago by SoyedTuhinAhmed
  1

• a little question about information entropy

  #12 opened 2 years ago by XA23i
  1

• 为什么在论文中不比较与ReActNet基于mobilenet-v1提出的新网络结构在应用ReCU后的效果呢?

  #2 opened 3 years ago by PeiqinSun
  1

• How does torch.clamp() reviving dead weights?

  #9 opened 3 years ago by YKrisLiu
  1

• Different shortcut design between cifar and imagenet in ResNet?

  #10 opened 3 years ago by kriskrisliu
  0

• 关于clamp

  #4 opened 3 years ago by WhereKey
  1

• When loading (official) pretrained weights it results in "unexpected kyes in state_dict" (which implies resnet architecture mismatches with pretrained weights)

  #8 opened 3 years ago by kriskrisliu
  2

• Why divide the std of activations only during training, while not in the inference mode?

  #6 opened 3 years ago by talenz
  3

• Why weights of intermedium layers are not binary(+1 or -1)?

  #7 opened 3 years ago by kriskrisliu
  1

• Q_tau 初始化是无穷大

  #3 opened 3 years ago by PeiqinSun
  5

• ReActNet architecture

  #5 opened 3 years ago by CosimoRulli
  1

• 论文里公式(12)

  #1 opened 4 years ago by songyonger
  2