Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



20 Commits

Repository files navigation

Instance-Batch Normalization Network


Xingang Pan, Ping Luo, Jianping Shi, Xiaoou Tang. "Two at Once: Enhancing Learning and Generalization Capacities via IBN-Net", ECCV2018.


  • IBN-Net is a CNN model with domain/appearance invariance. It carefully unifies instance normalization and batch normalization in a single deep network.
  • It provides a simple way to increase both modeling and generalization capacity without adding model complexity.
  • IBN-Net is especially suitable for cross domain or person/vehicle re-identification tasks, see michuanhaohao/reid-strong-baseline and strong baseline for ReID for more details.


  • Pytorch 0.4.1 or higher


Top1/Top5 error on the ImageNet validation set are reported. You may get different results when training your models with different random seed.

Model origin re-implementation IBN-Net
DenseNet-121 25.0/- 24.96/7.85 24.47/7.25 [pre-trained model]
DenseNet-169 23.6/- 24.02/7.06 23.25/6.51 [pre-trained model]
ResNet-18 - 30.24/10.92 29.17/10.24 [pre-trained model]
ResNet-34 - 26.70/8.58 25.78/8.19 [pre-trained model]
ResNet-50 24.7/7.8 24.27/7.08 22.54/6.32 [pre-trained model]
ResNet-101 23.6/7.1 22.48/6.23 21.39/5.59 [pre-trained model]
ResNeXt-101 21.2/5.6 21.31/5.74 20.88/5.42 [pre-trained model]
SE-ResNet-101 22.38/6.07 21.68/5.88 21.25/5.51 [pre-trained model]

The rank1/mAP on two Re-ID benchmarks Market1501 and DukeMTMC-reID (from michuanhaohao/reid-strong-baseline):

Backbone Market1501 DukeMTMC-reID
ResNet50 94.5 (85.9) 86.4 (76.4)
ResNet101 94.5 (87.1) 87.6 (77.6)
SeResNet50 94.4 (86.3) 86.4 (76.5)
SeResNet101 94.6 (87.3) 87.5 (78.0)
SeResNeXt50 94.9 (87.6) 88.0 (78.3)
SeResNeXt101 95.0 (88.0) 88.4 (79.0)
IBN-Net-a 95.0 (88.2) 90.1 (79.1)

Load IBN-Net from torch.hub

import torch
model = torch.hub.load('XingangPan/IBN-Net', 'resnet50_ibn_a', pretrained=True)

Testing/Training on ImageNet

  1. Clone the repository

    git clone
  2. Download ImageNet dataset (if you need to test or train on ImageNet). You may follow the instruction at fb.resnet.torch to process the validation set.


  1. Edit Modify model and data_path to yours.
    Options for model: resnet50_ibn_a, resnet50_ibn_b, resnet101_ibn_a, resnext101_ibn_a, se_resnet101_ibn_a, densenet121_ibn_a, densenet169_ibn_a.

  2. Run test script



  1. Edit Modify model and data_path to yours.
  2. Run train script


This code is developed based on bearpaw/pytorch-classification.

MXNet Implementation

Citing IBN-Net

  author = {Xingang Pan, Ping Luo, Jianping Shi, and Xiaoou Tang},  
  title = {Two at Once: Enhancing Learning and Generalization Capacities via IBN-Net},  
  booktitle = {ECCV},  
  year = {2018}