Open SkalskiP opened 1 year ago
@capjamesg, take a look at this issue. I'm not sure if this API change will somehow affect autodistill
.
Related issues to keep in mind:
@SkalskiP , can you handle this one? Some other things came up which needs my attention.
@SkalskiP , can you handle this one? Some other things came up which needs my attention.
If you don't mind I like to tackle this ?
@onuralpszr, go for it!
@onuralpszr still planing to pick it up?
I was already doing it. Will push pr soon
Just double checking as I want to include it in next release and it's going to happen next week: https://github.com/roboflow/supervision/discussions/310
@SkalskiP I'm interested in working on this. Will it be possible for you to assign it to me?
@SkalskiP I'm interested in working on this. Will it be possible for you to assign it to me?
For other changes I had to pause a bit but I already did work to be pushed. Sorry about that.
@onuralpszr, do you still want to be assigned to it?
@onuralpszr, do you still want to be assigned to it?
If you don't mind please
No worries ;)
@SkalskiP can I pick this up?
Hi @SkalskiP, is this issue still open?
Description
Migrate datasets
sv.ClassificationDataset.classes
andsv.DetectionDataset.classes
field fromList[str]
toDict[int, str]
. Necessary changes should be made in all.from_*
and.as_*
methods in bothsv.ClassificationDataset
andsv.DetectionDataset
.Use case
Using
list
to storeclasses
forces us to use consecutive and starting from 0 integers. In fact, formats such as COCO or PASCAL VOC and, more recently, YOLO useint: string
relationships to store classes. Our dependence onclasses
in the form oflist
forces us to unnecessarily mapclass_id
while loading the dataset.Testing
Make sure to create Google Colab where you install your version of Supervision and showcase a demo of the changes you made.