Three pixel error benchmark code

[tungngovn]

Thank you for your great work.

I have a question about the three pixel error benchmark in your code. Can you explain why the disp_true was less than 1? Should we replace the number 1 by the number 3 for the threshold? And why you consider the situation when disp_true < true_disp*0.05?

correct = (disp_true[index[0][:], index[1][:], index[2][:]] < 1)|(disp_true[index[0][:], index[1][:], index[2][:]] < true_disp[index[0][:], index[1][:], index[2][:]]*0.05)

This is code line 145 in file train.py

[devmentality]

Hi! Probably, this question is not relevant already:D Anyway I'll answer. This metrics seems to be the one that's used in KITTY benchmark http://www.cvlibs.net/datasets/kitti/eval_scene_flow.php?benchmark=stereo For this benchmark, we consider a pixel to be correctly estimated if the disparity or flow end-point error is <3px or <5%

[Zhaohuai-L]

I also want to know, in your code disp_truewas less than 1px instead of 3px, and I think the use of indexmay ncorrect, index[0] represents the coordinates of the first point and not the first dimension of disp_true. Also I don't know what your three_px_acc_all stands for? It is neither D1 nor bad1.0 or bad3.0