When I try reboflow’s model-assisted labeling feature, I find that when I use the pre-trained model for auto-labeling, it successfully generates some labels, but at the same time all my original labeled boxes disappear, which makes me confused.
How can I solve this problem, or is it possible that I can deploy two models in the model-assisted labeling function at the same time, so that I can add more classes to a training set, which would be helpful.
Hi Abraham,
Yes, this is a known interaction with the automated labeling and makes the tool less helpful for data that has annotations already applied. For this reason, to use model-assisted labeling you’ll want to make sure your model labels all objects you want to identify.
An alternative approach you could take would be to use the Annotate API - Roboflow which will give you labels and then you can go through the process of combining labels for your data. Combining annotations would be something you do outside of Roboflow.
Hi Abraham,
Excited to let you know that we now support adding annotations via Label Assist without removing previous annotations in your images. Let us know how it goes!
Hi @trevorhlynn Can you help me for resolve my Problem related to labelling when i annotate the image i did correctly labelling but when i give image to predict the sometimes coming correct but like 60 % coming wrong result labelling
Hi there! Yes, automated labeling is only as accurate as the model you’re using. You can adjust the confidence and overlap thresholds to adjust under what conditions a label will be applied. This can help reduce mislabeled objects.
If you’d like to use a more accurate model from Universe, you can do that by starring the project and then using it to apply labels in Annotate: Launch: Use Universe Models for Label Assist and Training