Skip to content
Train the HRNet model on ImageNet
Branch: master
Clone or download
Latest commit d698921 Apr 17, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
experiments add files Apr 10, 2019
lib fix a bug in Apr 10, 2019
tools Update Apr 17, 2019
LICENSE Update Apr 10, 2019

High-resolution networks (HRNets) for Image classification


This is the official code of high-resolution representations for ImageNet classification. We augment the HRNet with a classification head shown in the figure below. First, the four-resolution feature maps are fed into a bottleneck and the number of output channels are increased to 128, 256, 512, and 1024, respectively. Then, we downsample the high-resolution representations by a 2-strided 3x3 convolution outputting 256 channels and add them to the representations of the second-high-resolution representations. This process is repeated two times to get 1024 channels over the small resolution. Last, we transform 1024 channels to 2048 channels through a 1x1 convolution, followed by a global average pooling operation. The output 2048-dimensional representation is fed into the classifier.

ImageNet pretrained models

HRNetV2 ImageNet pretrained models are now available!

model #Params GFLOPs top-1 error top-5 error Link
HRNet-W18-C 21.3M 3.99 23.2% 6.6% OneDrive
HRNet-W30-C 37.7M 7.55 21.8% 5.8% OneDrive
HRNet-W32-C 41.2M 8.31 21.5% 5.8% OneDrive
HRNet-W40-C 57.6M 11.8 21.1% 5.5% OneDrive
HRNet-W44-C 67.1M 13.9 21.1% 5.6% OneDrive
HRNet-W48-C 77.5M 16.1 20.7% 5.5% OneDrive

Quick start


  1. Install PyTorch=0.4.1 following the official instructions
  2. git clone
  3. Install dependencies: pip install -r requirements.txt

Data preparation

You can follow the Pytorch implementation:

The data should be under ./data/imagenet/images/.

Train and test

Please specify the configuration file, the directory for data and the directory for saving log files and models.

bash --cfg <CONFIG-FILE> --dataDir <DATA-DIR> --logDir <LOG-DIR> --modelDir <MODEL-DIR>

For example, train the HRNet-W18 on ImageNet with a batch size of 128 on 4 GPUs:

python tools/ --cfg experiments/cls_hrnet_w18_sgd_lr5e-2_wd1e-4_bs32_x100.yaml --dataDir ./data/ --logDir ./log/ --modelDir ./output/

For example, test the HRNet-W18 on ImageNet on 4 GPUs:

python tools/ --cfg experiments/cls_hrnet_w18_sgd_lr5e-2_wd1e-4_bs32_x100.yaml --dataDir ./data/ --testModel hrnetv2_w18_imagenet_pretrained.pth

Other applications of HRNet


If you find this work or code is helpful in your research, please cite:

  title={Deep High-Resolution Representation Learning for Human Pose Estimation},
  author={Ke Sun and Bin Xiao and Dong Liu and Jingdong Wang},

  title={High-Resolution Representations for Labeling Pixels and Regions},
  author={Ke Sun and Yang Zhao and Borui Jiang and Tianheng Cheng and Bin Xiao 
  and Dong Liu and Yadong Mu and Xinggang Wang and Wenyu Liu and Jingdong Wang},
  journal   = {CoRR},
  volume    = {abs/1904.04514},


[1] Deep High-Resolution Representation Learning for Human Pose Estimation. Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. CVPR 2019. download

You can’t perform that action at this time.