Tuesday, February 21, 2017

2D to 3D Image Conversion using DMAG4

In order to better understand how Depth Map Automatic Generator 4 (DMAG4) (the semi-automatic 2d to 3d image conversion software) works, I wrote a little blurb about it, in particular, how I go about creating the sparse depth maps.

Here's the link: 2D to 3D Image Conversion using DMAG4.

Bonus wiggles/wobbles:


The singing butler.


It's a wonderful life.

Friday, February 17, 2017

Depth Map Improvement Using The Fast Bilateral Solver

I have written a technical report about depth map improvement using the fast bilateral solver of Jonathan Barron. It basically shows what Depth Map Automatic Generator 9b (DMAG9b) can do. Now, in that paper, I tried to improve the quality of depth maps generated by Depth Map Automatic Generator (DMAG), a notorious (but cool) creator of "fuzz" near object boundaries, so I might have asked too much. Still, DMAG9b is very good at making object boundaries crisp.

Here's the link: Depth Map Improvement Using The Fast Bilateral Solver.

Depth Map Generation using Graph Cuts

I have written a technical report about depth map generation using graph cuts which basically explains how Depth Map Automatic Generator 3 (DMAG3). The cool thing about DMAG3 is that it explicitly handles occluded pixels. This means that there is no need to compute a right depth map and perform a left-right consistency check. You can compare the depths maps obtained by DMAG3 in this technical report with depth maps obtained by DMAG2, DMAG5, DMAG6, and DMAG7 in A Qualitative Comparison of Selected Automatic DepthMap Generators.

Here's the link: Depth Map Generation using Graph Cuts.

Depth Map Generation using Optical Flow

I have written a technical report on depth map generation using optical flow which basically explains how Depth Map Automatic Generator (DMAG) works. As DMAG does not explicitly handle occluded pixels, it has a tendency to blur object boundaries. The cool thing about DMAG is that you don't have to give a minimum and maximum disparitiy. You can compare the depths maps obtained by DMAG in this technical report with depth maps obtained by DMAG2, DMAG5, DMAG6, and DMAG7 in A Qualitative Comparison of Selected Automatic DepthMap Generators.

Anyways, here's the link: Depth Map Generation using Optical Flow.

Thursday, February 16, 2017

Depth Map Automatic Generator 11b (DMAG11b)

DMAG11b is a tool that can be used to improve the quality of depth maps. Given a reference rgb image and an input depth map, DMAG11b detects pixels where the depth is unreliable and makes the depth at that pixel transparent in the output depth map. The output depth map can be fed to DMAG4, DMAG9, or DMAG11 (the programs used for 2d to 3d image conversion) to fill the transparent pixels for hopefully a better depth map than the one we started with.

To determine if the depth (coming from the input depth map) at a given pixel is reliable, DMAG11b computes the depth variance using the joint bilateral filter. A low (high) depth variance translates to a high (low, respectively) confidence. The output depth map is generated by making transparent any pixel for which the depth (coming from the input depth map) is of low confidence.

This idea of using the variance of the depth to compute a confidence comes from Jon Barron and his paper entitled "The Fast Bilateral Solver".

Here's an example:


Reference image.


Input depth map.


Depth confidence map. White (black) means high (low, respectively) confidence. This depth confidence map is computed internally by DMAG11b.


Output depth. Transparent pixels have the checkerboard pattern.

The windows executable (guaranteed to be virus free) is available for free via the 3D Software Page.

Monday, February 13, 2017

Depth Map Automatic Generator 11 (DMAG11)

DMAG11 is a joint bilateral filter which can be used to "densify" a sparse depth map in the context of 2d to 3d image conversion.

The input to DMAG11 (in terms of image files) is a reference image and a (sparse) depth map where where some depths have been scribbled upon. DMAG11 can be used to convert 2d images into 3d as an alternative to Depth Map Automatic Generator 4 (DMAG4) and Depth Map Automatic Generator 9 (DMAG9).

Here's an example:


Reference image.


Sparse depth map (input).


Dense depth map (output).

The windows executable (guaranteed to be virus free) is available for free via the 3D Software Page.