Deep Primal-Dual Networks

Datasets

The training data used for the depth only super-resolution method is available here. It consists of three tar files, one for each benchmark dataset used in our paper:

• dataset_mb.tar: noiseless Middleburry
• dataset_nmb.tar: noisy Middleburry
• dataset_tm_ta.tar: ToFMark

They mainly differ in the scaling of the depth values.

The training data for the guided depth super-resolution is available here.

Publications

The papers explaining the methods are on arxiv:

• ATGV-Net: Accurate Depth Super-Resolution
• A Deep Primal-Dual Network for Guided Depth Super-Resolution

If you find the code, or the data useful for your research, please cite

@inproceedings{riegler16dsr,
  title={ATGV-Net: Accurate Depth Super-Resolution},
  author={Riegler, Gernot and R\"{u}ther, Matthias and Bischof Horst},
  booktitle={European Conference on Computer Vision},
  year={2016}
}

@inproceedings{riegler16gdsr,
  title={A Deep Primal-Dual Network for Guided Depth Super-Resolution},
  author={Riegler, Gernot and Ferstl, David and R\"{u}ther, Matthias and Bischof Horst},
  booktitle={British Machine Vision Conference},
  year={2016}
}