I try to work with the Middlebury 2014 stereo dataset. The dataset contains ground truth disparity maps contained in a .pfm file.
From what I read online, I can load the .pfm file into a cv::Mat like so:
cv::Mat = cv::imread(path, cv::IMREAD_UNCHANGED);
This seems to correctly load the floating point values from the pfm. However, the per-pixel values don't make sense to me. They are much larger than the actual disparity values between left and right view.
So how do I convert those values to integral disparity values?