Missing Modules and Script Files for KITTI Depth Completion Reproduction

[JeonghwaHeo]

Dear Dr. Jiao,

Hi, thank you for sharing your code on GitHub.

I was trying to reproduce your work in my environment and encountered a few issues during the process. I hope you could clarify or provide some missing components.

1. Missing box_3d_encoder.py
   While running depth_completion_kitti.py, I encountered an import error related to box_3d_encoder. Since depth_completion_kitti.py imports obj_utils, and obj_utils.py further imports box_3d_encoder, it seems that the corresponding file is not included in the repository.
   Could you please provide the required box_3d_encoder implementation if it is available?

2. Missing MobileSAMv2 scripts
   In the “Generating Instance Masks” section of the README, the command bash ./experiments/mobilesamv2.sh is mentioned. However, the repository does not contain MobileSAMv2-related scripts or directories under /tools.
   Does your project use the vanilla MobileSAM?
   If possible, could you also upload the missing script files needed to reproduce your instance mask generation pipeline?

3. Depth input discrepancy (lidar_D80t vs lidar_D50t)
   In the “Training” step, train_kitti.py expects inputs from /lidar, /lidarSAM_t, and /lidar_D50t.
   However, depth_completion_kitti.py generates /lidar_D80t instead.
   Should the training be performed using lidar_D80t, or should it follow the default lidar_D50t input expected in train_kitti.py?

Thank you in advance for your help.

Any clarification or missing files would be greatly appreciated!

Best regards,
Jeonghwa Heo