Install

From URL: python -m pip install -U git+https://github.com/Soundbendor/yolo_dataset_tools.git#egg=yolo_dataset_tools

From local clone: python -m pip install -U PATH_TO_YOLO_DATASET_TOOLS

Interface files

dataset_merger.pyi dataset.pyi image.pyi annotation.pyi

Functions

1. Converts segmentation images => COCO JSON => YOLO segmentation format. (And, parses category list from file.)
2. Merges YOLO segmentation datasets by writing to new folder + storing metadata as JSON.

Function/class	Description
get_category_list_from_file	Parse list of category names from file.
write_partition_segmentation_images_to_yolo	Combines write_coco_json_from_segmentation_path, write_image_annotations_to_folder, and creation of partition folder.
write_coco_json_from_segmentation_path	Converts segmentation image masks to segmentation lists, storing them in COCO JSON format.
write_image_annotations_to_folder	COCO JSON => YOLO segmentation format.
DatasetMerger	Merge datasets.
Dataset	Access dataset items.

Each has a doc string and examples. (See main() or __main__ for examples.)

Purpose

• Some datasets save the labels as images themselves. YOLO needs them as a list of points, not as an image.
• write_coco_json_from_segmentation_path is an adapted version of an existing online script that converts from label images to COCO JSON format.
• From there, write_image_annotations_to_folder converts from COCO to YOLO.