How does it use on Object Detection dataset?

[monoloxo]

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

[monoloxo]

Object Detection dataset not this structure:

image_dir/ <dataset split>/ <label>/ <image

so,does it also work on object detection dataset?

[cfezequiel]

@monoloxo The default schema used by the tool is intended for an image classification dataset. But you should be able to repurpose it to generate TFRecord files for object detection. You can use the Flexible Schema to specify the fields that would contain the bounding boxes in addition to the image/label/split info, e.g.

split, image_uri, label, x0, y0, x1, y1, ...

You would also need to create the input CSV/DataFrame for TFRecorder that contains the necessary information.