Active Labeling Project

sellProject sellDFKI sellAugmented Vision sellDeep Learning sellActive Learning sellData Annotation

Active Learning for Optimal Data Labeling (Project)

Abstract

Topic

In this project, we will extend an existing dataset with new annotations using active learning. This involves training a neural network to predict the new annotations we want to add in the dataset, manually refining them in batches, and re-training the network. This network tells us which data samples to review in the next batch. This is an iterative process which is repeated until the whole dataset is done.

Tasks

  1. Create an initial dataset using pretrained models.
  2. Implement an active learning pipeline to aid in annotation.
  3. Manually review/refine predictions, finetune network, predict, repeat. This step involves significant amount of manual data annotation.

Expected Skills

  1. Python and PyTorch
  2. Basics of neural networks
  3. Familiarity with data annotation