Another key feature that you will be getting once a model has been defined as Current is the semi-automatic cleaning process.
Indeed, based on the sorting of images in gallery views, you are naturally proposed all the images for which the annotation and the prediction don't match. Those are the images you should look at in priority, as there may be an annotation error and you may want to modify it to improve your overall performance.
In the image above, we can see for a given concept the images sorted by the confidence score (on the left side for the first pages and on the right side for the last pages). The images with a red label have predictions that do not match with their annotation, and for most of them, we would indeed have to correct the annotation !
See Annotating Images to know how to correct the images.