About annotation

Community Help
1

Hello everybody,
I’ve seen that there are some changes in Roboflow.
The old version allowed to automatically divide the images between Train/Valid/Test Sets at the time of upload. This is no longer available.
I’ve seen that I have to annotate the images before I can transfer them to the dataset, but as far as I know the images in the Test Set don’t need to be annotated.
Could I have some help on my doubt? How do I manage the images that will go into the Test Set? I need to not annotate them for when I do the inference phase later.

Thanks a lot,
Emanuele

2

Hey there. You’ll want to annotate images for your Test Set as well otherwise you won’t be able to get metrics on how well your model is performing compared to ground truth data.

A dataset is currently defined as containing only annotated images. Having an unlabeled test set is not supported for the time-being (you may work around this by marking them as null).

That said, if this is a common thing that people want we’re happy to consider adding it. Please leave a :heart: on this message to register a vote for this feature request.