search

JiaRenChang / PSMNet

Pyramid Stereo Matching Network (CVPR2018)
MIT License
1.43k stars 424 forks source link

Some question about the result #144

Open Chenlei6 opened 5 years ago

Chenlei6 commented 5 years ago

In the training of the KITTI dataset, disparity estimation results will not be constrained or punished in some areas without Ground-Truth. When you are training, does the predicted result have outliers in the Ground-Truth-free area? If so, how did you solve it? I hope to hear from you.

WASCHMASCHINE commented 5 years ago

I cannot speak for the author, but I can confirm that there are strong artifacts for some parts of the scene that are not covered by the Lidar ground truth. For example, look at the Autobahn / highway scenes and notice the disparity estimates in the sky.

Chenlei6 commented 5 years ago

Thanks for your reply！