.download() re-downloads the same version even if it already exists on disk

[yeldarby]

trafficstars

The default behavior should be to use the local copy since versions are frozen in time. Perhaps we could provide a force=true option that deletes the existing one first to clear local changes.

But especially for large datasets this is wasting time and bandwidth when re-training.

Example from our YOLOv8 notebook:

[FrancescoSaverioZuppichini]

Hi @yeldarby :wave: , thanks for reporting it. I'll have a look :)

[FrancescoSaverioZuppichini]

@yeldarby note that the .download , accepts a parameteroverwrite that is set to True by default. Thus, everytime you downlod a dataset, it will overwrite your local one.

You can remove this behaviour by setting overwrite = False. Let me know if this helps

[yeldarby]

Should we make the default be False? When would you want to overwrite it? (Exported versions are immutable)

[capjamesg]

@yeldarby I am in support of this change. Want me to file a PR?

[capjamesg]

@yeldarby Please see https://github.com/roboflow/roboflow-python/pull/213.