Abstract
On this page, we provide datasets discussed in the paper: Synthezising Real World Stereo Challenges. With these datasets, we aim at isolating specific challenges for stereo matchers. Previous synthetic datasets did not seperate different problematic issues in stereo analysis.Datasets
We generated four datasets, addressing textureless areas, foreground fattening, decalibration and visual artifacts. To each dataset different levels (none, slight, strong) of white Gaussian noise are applied.
Textureless
A slanted planar surface, with diminishing texture towards the image center. 
Foreground fattening
Thin foreground objects. Background texture is diminishing towards the right image border. 
Decalibration
A gradient of vertical and horizontal texture. The match image is rotated slightly around the image center. 
Visual artifacts
A slanted planar surface. Blocks of different thickness and transparency are introduced to the left view. 
Ground truth
We provide ground truth as 16 bit graylevel pgm files. Disparities in pixel units are obtained by dividing the 16bit value by 256.
Dataset Download
We generated four datasets, addressing textureless areas, foreground fattening, decalibration and visual artifacts. To each dataset different levels (none, slight, strong) of white Gaussian noise are applied.
You can download the dataset via Zenodo here: https://zenodo.org/doi/10.5281/zenodo.8032801