ID 2434: Improving Handwriting Recognition with Data Generation

Symbolic picture for the article. The link opens the image in a large view.
Handwriting movements contain semantic infomation of content, which makes handwriting recognition possible by tracking the hand movenments of writers. This work aims to investigate text-conditional different data generative methods (GANs, diffusion model, etc.) for time-series data, and compare the performance of handwriting recognition model train with and without generated data.
The latest version of digipen of STABILO will be employed for this work. Although data of previous versions of digipen are provided, a data collection session will be included to evaluate the performance of proposed methods on the digipen.

Requirements

  • Strong background knowledge in deep learning, time-series processing and generative models.
  • Proficiency in Python and familiarity with PyTorch.
  • German skills are preferable for supporting in data collection.

Tasks

  • Review the existing literature on generative models for time-series data.
  • Implement text-conditional generative models to generate time-series data.
  • Collect data and evaluate model performance on latest version of digipen.

Supervisors

Jindong Li

Researcher & PhD Candidate

Please use the application form to apply for the topic. We will then get in contact with you.