No description or website provided.
Switch branches/tags
Nothing to show
Clone or download


This is the Caffe implementation of 3D Graph Neural Networks for RGBD Semantic Segmentation:



Required CUDA (7.0) + Ubuntu14.04.


For installation, please follow the instructions of Caffe and DeepLab v2.

Data Preparation

  1. Download the trained model (
  2. Download the prepared training data (prepared hdf5 data) (
  3. Download the testing data (
  4. Download the original provided data (


  1. Clone the repository.

  2. Build Caffe and matcaffe:

    cd caffe_code
    make -j8 && make matcaffe
  3. Inference:

    • Evaluation code is in folder 'matlabscript'.
    • Download trained models and unzip it. Pretrained model is saved in folder "model/nyu_40/".
    cd matlabscript
    run nyu_crop_data_mask_msc.m
    • The result is saved in folder "../result/nyu_40_msc/"
  4. Training:

    • Training data preparation
        cd matlabscript
        run generatedata (setting training = true)
        cd ..
        cd train_data_hdf5_file_generate
        python generate_hdf5
        cd ..

    We have also provided the training data in folder "traindata/"

    • Run caffe training


If you use our code for research, please cite our paper:

  title={3D Graph Neural Networks for RGBD Semantic Segmentation},
  author={Qi, Xiaojuan and Liao, Renjie and Jia, Jiaya and Fidler, Sanja and Urtasun, Raquel},


If you have any question or request about the code and data, please email me at . If you need more information for other datasets plesase send email.


MIT License