Exporting Annotations in a YOLO Format

[mwilson703]

Hi,

I'm wondering if species level annotations can be exported from biigle as a YOLO format?
Any guidance is super appreciated, thank you!!

[dlangenk]

Hi,

there is no official YOLO Biigle report, but you could export in Coco file format and then use the python fiftyone tool to convert to Yolo. Here is the command to do so for YoloV5 dataset:
'''fiftyone convert --input-type fiftyone.types.COCODetectionDataset --input-kwargs data-path= labels-path= --output-type fiftyone.types.dataset_types.YOLOv5Dataset --output-dir --output-kwargs split='''

If you run the command twice once with split=val and once with split=train but with the same output-dir, but two different coco input files it will not overwrite the dataset.yaml but just add the train and val datasets respectively.

[mwilson703]

Thank you so much! Is there a way to export in Coco file format from video annotations? For context, I'm wanting to work from the stills that are created from the annotations that I have made in the videos.

[dlangenk]

Unfortunately there is no option to do that at the moment. Your best bet would be to export the csv (only available report currently) and parse it to yolo. Or create a volume with the still images and use the api to export/import the annotations and then use the method described above. Both require some programming. Here you can find some examples of using the biigle api, as well as a python "library" to make working with the api easier. If you implement one of the two approaches, we would gladly appreciate a contribution to the community-resources as this could be helpful to other people as well.

[mwilson703]

This has been really helpful to getting my work rolling. I will absolutely contribute to the community-resources as my project develops further :)