search

sshan-zhao / GASDA

Geometry-Aware Symmetric Domain Adaptation for Monocular Depth Estimation, CVPR 2019
133 stars 23 forks source link

Regarding Ground truth for Kitti dataset #4

Closed koutilya-pnvr closed 5 years ago

koutilya-pnvr commented 5 years ago

https://github.com/sshan-zhao/GASDA/blob/2835031eae5beb8e26f7323c4c02e7643a830d2e/utils/dataset_util.py#L111

Why did you choose to build your own ground truth for Kitti dataset when you have the ground truth data available inherently in the official Kitti dataset.

sshan-zhao commented 5 years ago

https://github.com/sshan-zhao/GASDA/blob/2835031eae5beb8e26f7323c4c02e7643a830d2e/utils/dataset_util.py#L111

Why did you choose to build your own ground truth for Kitti dataset when you have the ground truth data available inherently in the official Kitti dataset.

Hi, Firstly, our evaluation strategy is the same as Godard et.al, and the training/validation/test splits might be different from the official. Secondly, actually the official ground truth is obtained by fusing the lidar data of multiple frames and correcting the depth via SGM, so I think the official ground truth is not quite different from the raw lidar data, but denser. As a result, I think it will not affect the final evaluation results. Thanks.

koutilya-pnvr commented 5 years ago

Fair enough. Thanks a lot for your response.