vacancy/Synchronized-BatchNorm-PyTorch

Issues

• test gap between training and test

  #30 opened 5 years ago by ZhiweiYan-96
  3

• spam

  #50 opened a year ago by QXPOMBA
  0

• .

  #49 opened 2 years ago by LSKhappychild
  0

• how to export with onnx

  #48 opened 2 years ago by tongchangD
  0

• Where is "track_running_stats" implementation code?

  #47 opened 3 years ago by junha1125
  1

• Training stuck with multiple call of forward function

  #46 opened 3 years ago by ArmastusChen
  7

• raining couldn 't start

  #45 opened 3 years ago by zwyking
  2

• Is it more difficult to train model?

  #18 opened 3 years ago by yuanfuqiang456
  4

• Training cannot start

  #40 opened 4 years ago by shuuchen
  7

• Train Stucked

  #44 opened 3 years ago by YinengXiong
  4

• a question about the highlight "use sqrt(max(var, eps)) instead of sqrt(var + eps)"

  #43 opened 3 years ago by louyj136
  4

• is everyone test the code in pytorch 1.0 ?

  #24 opened 3 years ago by wadesunyang
  10

• After sync batch norm is applied, more gpu memory is consumed

  #10 opened 6 years ago by shachoi
  9

• Is this a bug that channel between input tensor and sync batchnorm are mismatch the code still run successful?

  #31 opened 3 years ago by lscelory
  2

• module问题

  #39 opened 3 years ago by fanjing136
  0

• batchnor while using distributed dataparallel

  #42 opened 3 years ago by ParthaEth
  1

• Training cannot start

  #41 opened 4 years ago by shuuchen
  0

• Question on `sqrt(max(var, eps))`

  #38 opened 4 years ago by nmhkahn
  2

• How to use it when testing

  #33 opened 4 years ago by zhulanyun
  1

• Wired things, module must have its parameters and buffers on device cuda:0 (device_ids[0]) but found one of them on device: cpu

  #37 opened 4 years ago by kehuantiantang
  1

• Thinking about 'sync_batchnorm.batchnorm.convert_model(module)'..

  #35 opened 4 years ago by chrisway613
  1

• RuntimeError with convert_model - "found one of them on device: cpu"

  #34 opened 4 years ago by FreemanG
  2

• How to cite this repo in bib?

  #32 opened 5 years ago by Anikily
  0

• Does this support torch.nn.parallel.DistributedDataParallel?

  #1 opened 6 years ago by acgtyrant
  18

• Where is the "BatchNormReimpl"?

  #29 opened 5 years ago by skanderbug
  1

• Network performance is getting worse

  #28 opened 5 years ago by charlesCXK
  2

• about convert_model

  #26 opened 5 years ago by Re3write
  6

• Wired things happen when applied to FPN

  #3 opened 6 years ago by ShuLiu1993
  9

• Can't test successfully using the scripts in ./tests

  #20 opened 5 years ago by Yochengliu
  2

• Different from torch.nn.DataParallel that will separate total batch to each gpu equally?

  #23 opened 5 years ago by ksnzh
  1

• Can one past the test ?

  #17 opened 5 years ago by chenchr
  1

• shall I use your `DataParallelWithCallback` instead of pytorch `nn.DataParallel`?

  #19 opened 5 years ago by CoinCheung
  1

• About python test scripts and SyncBN usage

  #21 opened 5 years ago by Spritea
  1

• about fp16

  #22 opened 5 years ago by 666zz666
  5

• Unstable performance between adjacent epochs during test phase

  #5 opened 6 years ago by 7color94
  6

• Another test script for numerical stability

  #14 opened 6 years ago by acgtyrant
  7

• `np.fmax(npa, 1e-5))` does not consider that npa is negative

  #13 opened 6 years ago by acgtyrant
  2

• When I use the sync bn on 8x GPU, I will stop sometimes.

  #11 opened 6 years ago by ycszen
  2

• How to solve the numeric error?

  #12 opened 6 years ago by acgtyrant
  1

• How to load bias weights

  #7 opened 6 years ago by guantinglin
  1

• The mean is not averaged realy.

  #4 opened 6 years ago by acgtyrant
  1

• How to use this module to load the pre-trained model offered by the official PyTorch?

  #6 opened 6 years ago by ycszen
  3

• _SynchronizedBatchNorm's backward is redefined by its forward, right?

  #2 opened 6 years ago by acgtyrant
  1