Supplementary Material for Depth from Semi-Calibrated Stereo and Defocus

1. Rectification Result Comparison

In this part, we compare the disparity maps generated using our rectification, our optimization method, the depth upsampling method by Ferstl et al. [1], and the rectification method by Fusiello et al. [2].

Side View Image

Side View Disparity

Main View Image

Our Rectification

Our Method

Ferstl et al. [1]

Fusiello et al. [2]


















































2. Disparity Map Result Comparison

In this part, we compare the disparity maps generated using our method without exploiting the defocus cues, generated from using only the defocus cue, and the final results. For results generated using only defocus cue, after obtaining the disparities at edges, we propagate the results to the entire image using interpolation similar to Zhuo et al. [3]. We also compare with the the depth upsampling results by Ferstl et al. [1], the single image defocus maps by Zhuo et al. [3], the depth map from Lytro Illum, and the depth map from Kinect version 2 (for indoor scenes). Note that in some example (e.g. #10), the stereo completely fails due to large occlusion, while the defocus still works to give a reasonable result. For both indoor and outdoor, we show a fuzzy case at the end (#5 and #13) to demonstrate our ability to handle fuzzy objects.

Side View Image

Side View Disparity

Main View Image

Our Result (w/o defocus)

Our Defocus Result

Our Final Result

Ferstl et al. [1]

Zhuo et al. [3]

Lytro Illum

Kinect


















































































































[1] D. Ferstl, C. Reinbacher, R. Ranftl, M. R¨uther, and H. Bischof. Image guided depth upsampling using anisotropic total generalized variation. In International Conference on Computer Vision (ICCV), 2013.

[2] A. Fusiello and L. Irsara. Quasi-euclidean epipolar rectification of uncalibrated images. Machine Vision and Applications, 22(4):663–670, 2011.

[3] S. Zhuo and T. Sim. Defocus map estimation from a single image. Pattern Recognition, 44(9):1852–1858, 2011.