KITTI Dataset - minimum distance issue

[eduardohenriquearnold]

Thanks for sharing the source code of this paper.

I am exploring the dataset and I found a weird behaviour related to the minimum distance between poses in the KITTI Odometry dataset.

I observed that the translation vector norm is below the MIN_DIST threshold (following the paper, I used a threshold of 10m) for most of the samples. That means that most of the samples have relative distance below (but close to) the minimum distance of 10m.

To verify this behaviour one can simply inspect the value of pdist[cur_time, next_time] after

DeepGlobalRegistration/dataloader/kitti_loader.py

Line 278 in 46dd264

self.files.append((drive_id, curr_time, next_time))

I believe the problem is in line

DeepGlobalRegistration/dataloader/kitti_loader.py

Line 275 in 46dd264

next_time = next_time[0] + curr_time - 1

which should read

next_time = next_time[0] + curr_time

Can anyone else observe this behaviour?

[FanLu97]

@eduardohenriquearnold Hi, I have also observed this problem when I try to generate the KITTI odometry dataset. I am concerned that the issue may cause some degree of overfitting. Thus, I adopt a different method to process the dataset. I use the current frame with the 10th frame after it as a pair of point clouds to be aligned (this will result in a distance distribution from 5m to 15m, I think it is more variable). You can refer to our paper (https://arxiv.org/abs/2107.11992) and repo (https://github.com/ispc-lab/HRegNet) for the pre-computed KITTI odometry dataset.

[anhquancao]

Hi @eduardohenriquearnold, we fixed this minimum distance issue of DGR and reevaluate DGR under correct setting in our paper (https://arxiv.org/abs/2110.01269).
You can take a look at the fix in dataloader here:
https://github.com/valeoai/PCAM/blob/main/pcam/datasets/kitti_dataset.py#L59