Homography Propagation and Optimization for Wide-baseline Street Image Interpolation

Demo code for the work below:

• Nie, Y., Zhang, Z., Sun, H., Su, T., & Li, G. "Homography Propagation and Optimization for Wide-baseline Street Image Interpolation." IEEE Transactions on Visualization and Computer Graphics, 23(10), 2328-2341, 2017.

Video Demo on YouTube.

External libraries and code used:

• Boost 1.62: https://www.boost.org/
• OpenCV 2.49: https://opencv.org/

Run the code

1. The code is currently tested on VS 2012.
2. Configure the data path in the main.cpp file.
3. Define __FORWARD in nrdc_processing.h to run forward warping. The warping result will be saved in data/viewinterp/1.
4. Define __BACKWARD in nrdc_processing.h to run backward warping. The warping result will be saved in data/viewinterp/2. Note: DO not define __FORWARD and __BACKWARD at the same time.
5. Run the Matlab code in the folder multibandblending to render the final result, which will be save in data/viewterp/3.

Common Questions

Q1: How to generate the labels.txt file in the forward/backward folder?

A1: This file provides an initial segmentation of the image. Any reasonable superpixel segmentation algorithm (For example, SLIC) can be used.

Q2: How to generate the matches.txt file in the forward/backward folder?

A2: This file can be generated from any ANNF algorithms, for example, PatchMatch, DAISY Filter Flow and NRDC.

Citation:

@article{nie2017homography,

title={Homography propagation and optimization for wide-baseline street image interpolation},

author={Nie, Yongwei and Zhang, Zhensong and Sun, Hanqiu and Su, Tan and Li, Guiqing},

journal={IEEE transactions on visualization and computer graphics},

volume={23},

number={10},

pages={2328--2341},

year={2017},

publisher={IEEE} }

Code by Zhensong Zhang and Yongwei Nie. For research purpose ONLY.