haofeixu / aanet

[CVPR'20] AANet: Adaptive Aggregation Network for Efficient Stereo Matching
Apache License 2.0
528 stars 102 forks source link

Question about D1 metric #41

Closed LeoCho1learning closed 4 years ago

LeoCho1learning commented 4 years ago

Hi, thank you for your nice work. I have a question about the calculation of the d1 metric in your code. err_mask = (e > 3) & (e / d_gt > 0.05) (line 23 in metric.py)

In kitti benchmark, it says that a correctly estimated pixel should have the disparity error less than 3px or <5%.

But it seems that you are using 'and' instead of 'or'.

So I'm a littile confused, could you explain this?

haofeixu commented 4 years ago

Here we are measuring wrong pixels (i.e., > 3px and > 5%).