Skip to content
Go to file

Latest commit


Failed to load latest commit information.
Latest commit message
Commit time
Jun 18, 2017
Sep 9, 2019

ICNet for Real-Time Semantic Segmentation on High-Resolution Images

by Hengshuang Zhao, Xiaojuan Qi, Xiaoyong Shen, Jianping Shi, Jiaya Jia, details are in project page.


Based on PSPNet, this repository is build for evaluation in ICNet. For installation, please follow the description in PSPNet repository (support CUDA 7.0/7.5 + cuDNN v4).


  1. Clone the repository recursively:

    git clone --recursive
  2. Build Caffe and matcaffe:

    cp Makefile.config.example Makefile.config
    vim Makefile.config
    make -j8 && make matcaffe
    cd ..
  3. Evaluation mIoU:

    • Evaluation code is in folder 'evaluation'.

    • Download trained models and put them in folder 'evaluation/model':

      • icnet_cityscapes_train_30k.caffemodel: GoogleDrive

        (31M, md5: c7038630c4b6c869afaaadd811bdb539; train on trainset for 30k)

      • icnet_cityscapes_trainval_90k.caffemodel: GoogleDrive

        (31M, md5: 4f4dd9eecd465dd8de7e4cf88ba5d5d5; train on trainvalset for 90k)

    • Modify the related paths in 'eval_all.m':

      • Mainly variables 'data_root' and 'eval_list', and your image list for evaluation should be similar to that in folder 'evaluation/samplelist' if you use this evaluation code structure.
    cd evaluation
    vim eval_all.m
    • Run the evaluation scripts:
  4. Evaluation time:

    • To get inference time as accurate as possible, it's suggested to make sure the GPU card with specified ID in script '' is empty (without other processes executing)

    • Run the evaluation scripts:

  5. Results:

    • Prediction results will show in folder 'evaluation/mc_result' and the expected scores are:
      • ICNet train on trainset for 30K, evaluated on valset (mIoU/pAcc): 67.7/94.5
      • ICNet train on trainvalset for 90K, evaluated on testset (mIoU): 69.5
    • Log information of inference time will be in file 'time.log', approximately 33~36ms on TitanX.
  6. Demo video:

    • Video processed by ICNet on cityscapes dataset:
      • Alpha blending with value as 0.5: Video


If ICNet is useful for your research, please consider citing:

  title={ICNet for Real-Time Semantic Segmentation on High-Resolution Images},
  author={Zhao, Hengshuang and Qi, Xiaojuan and Shen, Xiaoyong and Shi, Jianping and Jia, Jiaya},


Please contact ''


ICNet for Real-Time Semantic Segmentation on High-Resolution Images, ECCV2018



No releases published
You can’t perform that action at this time.