Generating Point Cloud Pair Lists and Ground Truth Transformations for Custom Datasets

[hankeceli]

Hello,

I am currently experimenting with HRegNet using the Lyft Perception dataset. Given my novice understanding of point cloud registration, I apologize in advance for any basic questions.

In the README.md file, it is mentioned that:

The point cloud pairs list and the ground truth relative transformation are stored in data/kitti_list and data/nuscenes_list.

For my experiments with a Lyft dataset, I am uncertain about how to generate or compile the necessary txt files. Could you please guide me on the process of creating these txt files for a custom dataset?

I appreciate your assistance.

Thank you.

[FanLu97]

Hi, thanks for your interests.
We only require the source point cloud, target point cloud and the relative transformation (R, t) from source point cloud to target point cloud. Each line of the txt file are organized as "[source file name] [target file name] [12 dim transformation matrix (3x4)]". You can organize your data according the format. You may modify the dataloader (data/kitti_data.py or data/nuscene_data.py) to load your own data.