clovaai/overhaul-distillation

Issues

• ResNet18/34

  #40 opened 2 years ago by gadhane
  2

• ImageNet Trained model

  #39 opened 2 years ago by gadhane
  2

• Saving the Distilled Model

  #38 opened 2 years ago by gadhane
  4

• A new implementation for overhual distillation

  #36 opened 2 years ago by zjykzj
  1

• Does t_net also need to be trained?

  #37 opened 2 years ago by TouchSkyWf
  1

• torch.no_grad() is impotant?

  #34 opened 2 years ago by sakh251
  1

• About extracting pre relu features

  #35 opened 3 years ago by RuiFeiHe
  2

• About the margin

  #32 opened 4 years ago by Fangkang515
  1

• student model every layer no need relu activation funcution in validation,is it ok?

  #33 opened 4 years ago by jerrytony
  0

• Question about how to calculate m in margin ReLU

  #31 opened 4 years ago by Nusselder9
  2

• detection?

  #20 opened 4 years ago by jianbaba
  3

• KD with Resnet and Mobinetv3

  #30 opened 4 years ago by hamhanry
  4

• how to compute margin from BN statistics?

  #29 opened 4 years ago by Pixie8888
  1

• A question about equation 4

  #28 opened 4 years ago by peter943
  2

• why?????loss = loss_CE + loss_distill.sum() / batch_size / 10000

  #12 opened 5 years ago by jianbaba
  7

• loss 계산하는 과정에서 1000 으로 나누어주는 부분에 대해서 질문드립니다.

  #27 opened 4 years ago by cbpark-nota
  2

• The "freeze_support()" line can be omitted if the program is not going to be frozen to produce an executable.

  #26 opened 4 years ago by musab-r
  1

• low speed when training the network

  #24 opened 4 years ago by PkuDavidGuan
  2

• about measure result about model size

  #25 opened 4 years ago by ZainZhao
  2

• question about get_margin_from_BN function

  #19 opened 4 years ago by jiseo
  1

• Implementation of other methods

  #17 opened 4 years ago by GUOShuxuan
  2

• question abount margin seting

  #6 opened 4 years ago by SaintLogos1234
  3

• Pretrained models for CIFAR10

  #22 opened 4 years ago by zhangman164
  3

• 1

  #21 opened 5 years ago by ooodragon94
  0

• how to set the alpha parameters?

  #2 opened 5 years ago by ray-lee-94
  4

• Have your tried configuration where teacher and student are totally differet, such as densenet and resnet, or densenet and mobilenet?

  #16 opened 5 years ago by leoozy
  1

• Why you have to use a regressor for the teacher model?

  #14 opened 5 years ago by leoozy
  3

• Normalization factor Z in margin value

  #18 opened 5 years ago by winycg
  2

• What is the "Z" in your equa 10

  #15 opened 5 years ago by leoozy
  1

• bad performance for setting resnet20 as a student and resnet56 as a teacher

  #13 opened 5 years ago by Slawlight
  8

• A question about Experimental results

  #9 opened 5 years ago by baek85
  1

• Training configuration about teacher WRN28-4.

  #11 opened 5 years ago by Abyssaledge
  3

• Semantic segmentation source

  #1 opened 5 years ago by myth01
  2

• Question about loss_distill

  #10 opened 5 years ago by Sundrops
  5

• Question about the change of the size of feature

  #8 opened 5 years ago by Ernie1
  2

• How to calculate Mc without BatchNorm Layer

  #7 opened 5 years ago by hwc1824
  4

• ResNet architecture

  #5 opened 5 years ago by mileyan
  2

• about ResNet.py

  #4 opened 5 years ago by Wondersui
  0

• train student and set alpha=0, the valid_loss goes up and acc drop a lot

  #3 opened 5 years ago by ray-lee-94
  0