• SurfaceNet+ is available in 2020 TPAMI!

M. Ji, J. Gall, H. Zheng, Y. Liu, and L. Fang. SurfaceNet: An End-to-end 3D Neural Network for Multiview Stereopsis. ICCV, 2017

The poster pdf is also available.

1. install Nvidia driver 375 + cuda 8.0 + cudnn v5.1
2. install the conda environment by: bash installEnv.sh
   • DON'T WORRY, conda will generate an isolated environment for SurfaceNet with python2.7, anaconda, theano, ... etc. That means all your libraries / packeges' version will not be affacted, at the same time the ~/.bashrc file will not be changed.
   • before you run, PLEASE change the CUDA/CUDNN path in the files:
     • ./config/activate-cuda.sh change the 1st line to your cuda path, e.g.: export CUDA_ROOT=/usr/local/cuda
     • ./config/activate-cudnn.sh change the 1st line to your cudnn path, e.g.: export CUDNN_ROOT=/home/<your-user-name>/libs/cudnn
3. download the network model to the folder "./inputs/SurfaceNet_models" from the Dropbox folder
4. if the conda environment has been installed, one can activate it by: . activate SurfaceNet; deactivate it by: . deactivate.
5. in terminal run: python main.py

Some evaluation results are uploaded, including '.ply' files and the detailed number of Table 3. This could be helpful if you want to compare with this work.

SurfaceNet is released under the MIT License (refer to the LICENSE file for details).

If you find SurfaceNet useful in your research, please consider citing:

@inproceedings{ji2017surfacenet,
  title={SurfaceNet: An End-To-End 3D Neural Network for Multiview Stereopsis},
  author={Ji, Mengqi and Gall, Juergen and Zheng, Haitian and Liu, Yebin and Fang, Lu},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision (ICCV)},
  pages={2307--2315},
  year={2017}
}