Skip to content

Training flashocc-r50.py on custom dataset: decent near-range fitting but severe far-range missed detections #109

@EASONL00000

Description

@EASONL00000

Why is the performance poor when I train the flashocc-r50.py configuration on my own dataset? The near-range parts fit reasonably well on the test set, but the far-range parts are mostly missed. The training set also shows missed detections at far distances. The data format is aligned with nuScenes, with an occupancy grid size of 40 cm, 40k frames for training, and 4k frames for testing. The visualization results are very poor.

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