Export label empty files - yolov8 export confusion

cvat-ai / cvat

Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.

https://cvat.ai

MIT License

12.62k stars 3.01k forks source link

Export label empty files - yolov8 export confusion #8367

Closed davy-blavette closed 2 months ago

davy-blavette commented 2 months ago

Actions before raising this issue

[X] I searched the existing issues and did not find anything similar.
[X] I read/searched the docs

Steps to Reproduce

We don't have the tags in the exports. I have tried all versions of the tags: https://docs.cvat.ai/docs/manual/advanced/formats/#data-export-formats

the generated .txt label files have no value. yet my jobs are completed.

Expected Behavior

export with dashboard or sdk but the same problem.

Possible Solution

No response

Context

No response

Environment

No response

bsekachev commented 2 months ago

Which formats do not export tags (but expected to export them according to the documentation)?

davy-blavette commented 2 months ago

Its ok sry, it was my mistake

davy-blavette commented 2 months ago

hum, I tested the two latest export formats YOLOv8 Detection 1.0 and YOLOv8 Oriented Bounding Boxes 1.0 but the label files were empty before I used COCO 1.0 (which still works). I have a dataset with maskshape (Oriented Bounding Boxes) for detection.

it works with YOLOv8 Segmentation 1.0...it's a bit strange, the export format depends on our annotation and not on what we want to export. You can mask for detection, can't you ^^ ? And for rectangle shape it's ok ...

Rectangle shape it's 0 0.495214 0.500514 0.981571 0.269829
MaskShape 0 0.421875 0.443750 0.417187 0.445312 0.414062 0.448437 0.412500 0.448437 0.410938 etc.

it would be simpler to keep a generic format, and choose via check box, rectangle, shape or only Class...For manage all format

and for yolov8 classification, I'm still using ImageNet 1.0 for tags but can't find the right export among the 4 :)

bsekachev commented 2 months ago

If you want to export everything (or almost everything), please, use CVAT or Datumaro formats.

I have a dataset with maskshape (Oriented Bounding Boxes) for detection ...... works with YOLOv8 Segmentation 1.0

Oriented bounding box may be easliy converted to a mask. That is why we support such conversion.

We won't add support for an object to a format if specification of that format does not support such objects. For example YOLOv8 does not have unified format structure for all shapes. That is why we devided it to several formats.