Skip to content

Image to labels validation functionality #190

@merryHunter

Description

@merryHunter

Hi,

I am wondering, do you think it makes sense to introduce validation of image paths for annotations? like pyodi validate trainval.json --data-dir $DATA so that it outputs if there are any missing images while annotations for them do exist. Let me explain briefly the reasoning.

In my case, I often have issues with exporting datasets in COCO format from CVAT. Suppose I have a few projects where by some reason images have same name. Then, after I bring down images to a single folder, some images will be overwritten, or maybe they were missed from the beginning. Then, annotations may exist, but there is no easy out of the box tool to check the correspondence between annotations and images (particularly after all merge/random split operations).

Could it make sense to eliminate annotations for missing images automatically?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions