Difference b/w KITTI and KITTI-with-improved-GT

[surajiitd]

Thank you for making your code public ! I saw the results in your paper in Table 1 and 2. What is the difference b/w "KITTI eigen benchmark" and "KITTI-with-improved-groundtruth" ? What do you call these depths that are on the official KITTI website , and where can I find the other one?

[hisfog]

You can run the export_gt_depth.py to get gt_depths of KITTI eigen benchmark and KITTI-with-improved-groundtruth.

python export_gt_depth.py --data_path kitti_data --split eigen
python export_gt_depth.py --data_path kitti_data --split eigen_benchmark

Depths on the official KITTI website is another test split, neither eigen nor eigen_benchmark, and the gt_depths are not publicly available. See https://github.com/nianticlabs/monodepth2#-kitti-evaluation

[surajiitd]

Okay, the groundtruth depths are same. Just the train-test splits are different. Thank you @hisfog for clarification!!