[BUG] YOLOv5 yaml path ignored

[victor1cea]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 22.04
• FiftyOne installed from (pip or source): pip
• FiftyOne version (run fiftyone --version): 0.16.3
• Python version: 3.9.12

Describe the problem

The 'path' from dataset.yaml in YOLOv5 datasets is ignored/not handled.
Example of dataset.yaml stored

names:
[cars]
nc: 1
train: ./images/train
val: ./images/val
test: ./images/test
path: some/path # ignored

path should be used like a base path for the splits e.g. train path would actually be some/path/images/train. The importer ignores this and the splits are not found. The exporter does not write path: ... in the exported dataset.yaml.

Code to reproduce issue

Provide a reproducible test case that is the bare minimum necessary to generate
the problem.

# read any yolov5 and have values for "path" key in dataset.yaml
yolo_dataset: fo.Dataset = fo.Dataset(name)
for split in splits:
    yolo_dataset.add_dir(
        yaml_path=yaml_path,
        dataset_type=fo.types.YOLOv5Dataset,
        split=split,
        tags=split,
        include_all_data=True
    )

# export the dataset
split_view = self.yolo_dataset.match_tags(split)
split_view.export(export_dir,
                  yaml_path=export_dir + '/dataset.yaml',
                  dataset_type=fo.types.YOLOv5Dataset,
                  label_field='ground_truth',
                  split=split,
                  include_confidence=True
                  )

Other info / logs

What areas of FiftyOne does this bug affect?

• App: FiftyOne application issue
• Core: Core fiftyone Python library issue
• Server: Fiftyone server issue

Willingness to contribute

The FiftyOne Community encourages bug fix contributions. Would you or another
member of your organization be willing to contribute a fix for this bug to the
FiftyOne codebase?

• Yes. I can contribute a fix for this bug independently.
• Yes. I would be willing to contribute a fix for this bug with guidance
  from the FiftyOne community.
• No. I cannot contribute a bug fix at this time.

[brimoor]

Hi @victor1cea 👋

Feel free to make a quick PR to add support for an optional path parameter; I wasn't aware that it was a possible parameter in dataset.yaml. Do you have a link for reference?

It'd be great if you could update the link in our YOLOv5 documentation to point to a place that provides more complete documentation for dataset.yaml 🤗

https://github.com/voxel51/fiftyone/blob/develop/docs/source/user_guide/dataset_creation/datasets.rst#yolov5dataset
https://github.com/voxel51/fiftyone/blob/develop/docs/source/user_guide/export_datasets.rst#yolov5dataset

[victor1cea]

Hi @brimoor
I will gladly make a PR.
This is the reference from the official repo.
Go to 1. Create Dataset -> Or manually prepare your dataset -> 1.1 Create dataset.yaml.
They also unofficially support the optional download parameter as seen here. If you think that support for this parameter is needed, just let me know.

[brimoor]

Cool! Supporting download is fine with me if you'd like to add it. I guess it would download the zip and unpack it to the locations specified by train, val, path, etc.?

[victor1cea]

Yes, exactly. I will look into adding support for download too. 😄

[brimoor]

Fixed by #2016!