This repository contains the source code corresponding to the method presented in the paper:

• Confidence driven TGV fusion, V. Ntouskos, F. Pirri, arXiv: arXiv:1603.09302, 2016.

###Contact:

Valsamis Ntouskos [email protected], http://www.diag.uniroma1.it/~ntouskos;

ALCOR Lab [email protected] http://www.diag.uniroma1.it/~alcor,

###Instructions###

To run the demo you need to download the KITTI 2012 multiview dataset (data, calibration files, and multi-view extension) from here.

The disparity maps for all the stereo-pairs of the dataset, computed based on the method presented in the paper "Confidence driven TGV fusion", J. Zbontar and Y. LeCun, CVPR, 2015, are available for download from here.

###Quickstart###

1. Compile the CUDA kernels located in the ./cuda folder. Under Windows you can use the provided Makefile: nmake -f Makefile.win all;

2. Modify the path variables basedir and w_path in the demo script demo_kitti.

3. Execute the demo script: demo_kitti.

##Synthetic Datasets ###Instructions### You can run the scripts that process the synthetic datasets used in the paper. These datasets are available for download from here.

###Quickstart###

1. Compile the CUDA kernels located in the ./cuda folder. Under Windows you can use the provided Makefile: nmake -f Makefile.win all;

2. Modify the path variables basedir and w_path in the demo scripts proc_synth_data_objects and proc_synth_data_urban.

3. Execute the script: proc_synth_data_objects or proc_synth_data_urban.

4. Use script showSurf to visualize the surfaces corresponding to the fused depth images.

You need the CUDA Toolkit (version 6 or later) and a compatible compiler in order to compile the CUDA kernels.