KIHT models

KIHT project

Presentation

The aim of the KIHT project is to develop an intelligent learning device (« DigiPen » electronic pen) for digitized handwriting. The electronic « DigiPen » hardware is designed by STABILO with the help of the German Karlsruhe Institute of Technology (KIT).

Digital devices can help pupils and teachers in the learning process by promoting active learning techniques and providing immediate feedbacks. The e-learning literature shows that computer-based analysis of handwriting can be really accurate, sensitive, and reliable to produce relevant and consistent feedbacks for correction or guidance.

The models designed during this project are dedicated to handwriting trace reconstruction. Input of models comes from IMU sensors embedded in the Digipen and the ground truth is the writing trace on a tablet acquired during a dual acquisition process.

Related links

Conditions of Use

1. Purpose and Scope

  • 1.1 The models are provided for research purposes only.
  • 1.2 Users must agree to use the models solely for academic, educational, or scientific research. Commercial use is strictly prohibited unless explicitly authorized in writing.

Models

KIHT TCN-based models

Several models are given for comparison with benchmark results:

  • TCN: a TCN-based model, as in [1]
  • SE-TCN: a TCN-based model including a Squeeze and Excitation block, as in [In a paper in submission]
  • Dual-TCN: a TCN-based model with two specific heads, as in [In a paper in submission]
  • Dual-SE-TCN: a TCN-based model including a Squeeze and Excitation block and two specific heads, as in [In a paper in submission].

Note that the Mixture-Of-Experts model presented in [2] can be done by combining two TCN models from [1], one expert trained only on strokes (pen-down, without pen-up) and one expert trained on the full sequences (including strokes and pen-up motions).

References

If you use the model, you agree to cite the references related to the article:

[1] Swaileh, W., Imbert, F., Soullard, Y. et al. Online handwriting trajectory reconstruction from kinematic sensors using temporal convolutional network. IJDAR (2023).

[2] Florent Imbert, Eric Anquetil, Yann Soullard, Romain Tavenard. Mixture-of-experts for handwriting trajectory reconstruction from IMU sensors. Pattern Recognition, 2024, 161, pp.111231. ⟨10.1016/j.patcog.2024.111231⟩. ⟨hal-04811975⟩.

Download link

Before downloading the model, you agree that this model is under the CLIC licence and can only be used for research purposes. To receive the download link, please complete the following contact form.