Computational 3D Imaging with Position Sensors

Code and data for "Computational 3D Imaging with Position Sensors" in ICCV 2023.

Code

We include a physically-accurate two-bounce rendered implemented in MATLAB in two_bounce/. macro_scan.m demonstrates how to perform a 3D line scan of an object using with and without global illumination suppression. macro_nPatt_v_sc.m demonstrates how to sweep the number of patterns and pattern scale.

Data

3D point clouds from our lab prototype are in point_clouds as .ply files. They can be viewed in MeshLab.

-no-suppression denotes a single raster scan was used.

-minmax denotes the min/max processing of Nayar et al. [1] was used for global illumination suppression.

-ours denotes our proposed regression method was used for global illumination suppression.

Each point cloud is post-processed with bilateral filtering on the depth map, and points whose total intensity is below a threshold are removed.

Paper

Supplemental

Citation

@inproceedings{klotz2023psd3d,
 author = {Klotz, Jeremy and Gupta, Mohit and Sankaranarayanan, Aswin C.},
 title = {Computational 3D Imaging with Position Detectors},
 booktitle = {IEEE Intl. Conf. Computer Vision (ICCV)},
 year = {2023},
}

References

[1]: S. K. Nayar, G. Krishnan, M. D. Grossberg, and R. Raskar. Fast separation of direct and global components of a scene using high frequency illumination. In ACM Transactions on Graphics, volume 25, pages 935–944. 2006