Why running twice for MiddleBury

[kyuhyoung]

In the file 'test_rob.py', the network is applied two times for Middlebury images.
And it says "Run caffe for Middlebury2014 one more time to calculate right disparity".
What is this for and why is this needed only for Middelbury images?

[leonzfa]

@kyuhyoung When you submit your results to Middlebury benchmark, you should provide both left disparity and right disparity. This is only required by Middlebury benchmark. For KITTI and ETH3D, only left disparity is needed.