Skip to content

How to reproduce the results on the mulran dataset #9

@neng-wang

Description

@neng-wang

Hello @Pamphlett ,

I recently attempted to reproduce Outram's results on the MulRan dataset. However, the test outcomes showed significant translation and rotation errors, as follows:

Image

This is this results generated from Kaist01:1561000444390857630.bin.

I suspect the issue might lie in certain coordinate transformations between Kaist01 and the semantic_cluster_map (Kaist02) (i.e This ground truth pose may be inconsistent with the one processed in the semantic_cluster_map, so the gt_lidar poses I generated in the figure above should be inconsistent with your results). The semantic nodes appear to be correctly matched to the map. Here’s what I did:

  • I built the database using the Kaist02 semantic_cluster_map provided in [issue#6], with semantic labels inferred by SPVNAS.

  • Following the clues in [Issue#7], I subtracted the corresponding values from the ground truth pose (x and y coordinates) of each Kaist01 LiDAR scan.

Image
  • The calibration matrix provided in global_localization.yaml(calibration_matrix-Mulran) was applied.

Why does this result occur? Is there something wrong with my configuration?
Additionally, are the Outram parameter settings the same for the MulRan dataset as for the MCD dataset?

I’d greatly appreciate any insights you might have. Looking forward to your reply.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions