I have two pictures of the same object, taken by a car travelling down the road like shown on the right side of the image below. I want to find pixels of the object in each frame that correspond to each other.
Now, in the description of the Middlebury Stereo Evaluation v.3 dataset it says
Maximum disparities range from 200 to 800 pixels at full resolution.
This leads me to my two questions:
- Do I understand correctly that algorithms working with the Middlebury dataset had to match pixels that were a distance of 200 to 800 pixels apart, like shown on the left side of the image?
- Consider the leaderboard for the Middlebury Stereo Evaluation. Does an average absolute error metric of 1.4 mean, that above problem could be solved for the images in the dataset with an average accuracy of 1.4 pixels?