MVCMv0
Multiview Confidence Merging Version Zero
CPU-only implementation of SGBM inspired stereo matching engine applied to horizontal and vertical baselines. Results get shifted into the center view and merged using the shifted confidence measures in a simple winner-takes-all approach.
The disparity calculation is sliced and all slices for both baselines are calculated in parallel
Uses only four views: top/bottom/left/right view
only tunable parameter is the number of threads = number of slices to use; Submission uses 4 Slices (2 per Image)
SAD Aggregation Size: 9x9
Disparity scaling exponent: 6
all other parameters are kept at defaults ( no texture filtering, no speckle-removal);
initial offset for compare images of 33 pixels is used to have only positive disparities.
C++
Usage of four x64 Cores (i7-5600U @2.6GHz, 2Cores with Hyperthreading, 16GB RAM) on Dell Laptop (Windows 7)
