Looking for recommendations on re-labeling a large number of images in my dataset

Community Help
1

I’m looking at the Confusion Matrix on the latest version of my project and realizing part of the problem with my False Negatives is that a whole portion of my dataset is mis-labeled, but the majority of it is fine. Aside from exporting it, running a script to re-label the images and then uploading this back into Roboflow as a separate dataset, what are some of the suggested approaches to this ?
Thanks!

2

Hi @espressobot - if one class is consistently labeled as another, you can use our class remapping functionality (on the class tab).

You could also tag the impacted images and then create a version of only that tag and then remap just those classes (using the remap classes and filter by tag preprocessing steps). You’d then re-upload those images in a new project with your other images.

3

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.