Question:
How to accept or reject model predictions as annotations with out a ML backend (just UI, and API/SDK)?
I’m working with images and bounding boxes. I’m hoping to use this as iterative pre-annotations and manually review the model results, then feed them back for more training.
Setup: I’ve manually uploaded predictions through the python SDK. These are bounding boxes generated via YOLO. Unfortunately external constraints force me to run the model and upload results manually instead of linking to the backend. I’m using local storage currently.
I now have tasks, each with one or more predictions. Each prediction shows up in a separate tab (kind of annoying, I wasn’t able to figure out how to make them appear in one tab when uploaded through the sdk). I’m using this sdk command in a loop for each box result: client.predictions.create(task=task, **prediction) where **prediction is all the needed keywords. This works ok.
I can see the predictions, but how do I accept or reject them one-by-one to convert to annotations? I read about this with pre-annotations in the docs. I’ve seen a command to convert all of these to annotations but that’s not the desire, I want to go one-by-one in the UI to do QC on model outputs. I feel like I’m missing a button in the UI that should be readily apparent.
Version: 1.14.0.post0
Installed through pip
Thank you in advance!