Hi,
I’m looking for the best way to setup my projects such that I can have object detection model find the boxes of different categories, which I can then isolate and classify using a single label classifcation model.
My flawed way of doing this right now is having a object detection project where i annotate everything. My annotations follow the structure category-type1-type2-class. E.g. RB-F-T-1. For now I’m only concerned with finding category and class (type models might become relevant later).
I then generate a version with all boxes isolated and import it into a single label image classification project.
- When generating dataset versions for object detection models i modify all classes to only contain the category.
- When generating dataset versions for classification (single label) i modify all classes to only contain a distinct category and then it’s class (category-class).
I usually train classifcation models in Roboflow and the confusion matrix feature exposes bad annotations, which I want to alter. But if I change the single label annotation in my classification project (containing the isolated box image) i would like the bbox annotation in my object detection project to be updated aswell. Is there a way to go about this? Am I overthinking it? Should I restructure my projects and the way i annotate?
Hoping someone can make me wiser with their expertise:)
Thanks in advance!