Division of Microelectronic Systems Design (EMS)

Under way

Exploration of Deep Neural Network Topologies and Efficient Hardware Architecture for Historical Document Baseline Detection

Type of work:

Master Thesis / Diplomarbeit

Assignment:

In this work, a student has to search literature for the state-of-the-art deep learning based methods for baseline detection of historical handwritten documents. As the next step, the method with the best trade-off between complexity and accuracy has to be implemented in PyTorch using Python. A student has to explore trade-offs between complexity and accuracy for the implemented model using complexity reduction techniques. A hardware architecture has to be designed and implemented using Xilinx Vivado tools.

Skills:

  • Interest in Deep Learning and FPGAs
  • Experience with Python and C/C++
  • Experience with PyTorch and Xilinx Vivado HLS

Background:

Baseline detection of historical documents is an important step in historical documents recognition chain that is used to digitize historical documents and preserve the knowledge from disappearance.

Supervisor:

V. Rybalkin

Student:

Ahmed EL-Yamany

Year:

2020

Zum Seitenanfang