Description

Converts a dataset's labels from a format to another. Currently, the available format wrappers are:

• Yolo
• VOC
• Coco (Only one label per image)

Syntax

dataset-converter <INPUT_PATH> <OUTPUT_PATH> <INPUT_WRAPPER> <OUTPUT_WRAPPER> [--filter-classes (-f) <CLASS> [<CLASS> ...]] [--no-img (-n)]

Arguments

<INPUT_PATH> Input directory with the images and their labels.

<OUTPUT_PATH> Output directory that will contain all converted labels and their images (if there isn't --no-img).

<INPUT_WRAPPER> Input Wrapper.

<OUTPUT_WRAPPER> Output Wrapper.

[--filter-classes (-f) <CLASS> [<CLASS> ...]] List of classes to be filtered. All images that don't contain at least one of these classes will be ignored. - filtering is disabled by default

[--no-img (-n)] Don't copy images to <OUTPUT_PATH>.

Usage

converter = DatasetConverter()
converter.convert(
    in_dir, out_dir, in_wrap, out_wrap, [classes], [copy], [recursive]
)

in_dir and out_dir are both directory paths

in_wrap and out_wrap are both strings

classes is a list

copy and recursive are booleans