0

I have created a labeling job using AWS Ground Truth with the following details:

  • Task type: bounding boxes

  • Dataset type: images

  • Total dataset objects: 36,000

Objective: I would like to create a training job from the so far labeled data (~5000/36,000)

Issue: The 'LabelingJobOutput' hasn't been generated because the labeling job is not complete so I cannot make use of this notebook that I'm following.

Is there a workaround this?

iamarchisha
  • 175
  • 7
  • From reading the unclear and distributed AWS documentation along with other resources, I do not think it is possible to create a training job from 'in-progress' or 'stopped' labeling job. – iamarchisha May 03 '21 at 09:02
  • You have two options: 1) Consolidate workers responses in `consolidated-annotations` into intermediate output manifest; 2) (preferred option) implement [Active Learning Loop](https://aws.amazon.com/blogs/machine-learning/bring-your-own-model-for-amazon-sagemaker-labeling-workflows-with-active-learning/) where you cant set number of objects for each loop. – wts May 14 '21 at 15:04

0 Answers0