ultralytics.data.utils.HUBDatasetStats always makes new directory about hub HOT 5 CLOSED

@Burhan-Q @kalenmike opened ultralytics/ultralytics#8334 with a fix

from hub.

@kalenmike the class is used in ultralytics.hub.check_dataset which is referenced in the Docs here as a means to verify the dataset before uploading to HUB. The reason I raised this issue is because when I ran the function, it would create this directory without any data and I would have to remove the directory each time. Thought it would be better to move this so it wasn't called every time the class was initialized.

from hub.

Yes that's correct also, the same function is used in both places, this way we guarantee that if it works locally it should also work on the server

from hub.

@Burhan-Q This is an internal method used for HUB. It is not intended to be used outside of HUB. We use it to scan the dataset and generate the preview images. These images are then saved for rendering on the HUB in lower resolution and a more optimized size.

from hub.

@Burhan-Q yeah I think your understanding is right, perhaps the -hub dir is not always needed. I'll create a PR in ultralytics to update the behavior.

from hub.