Bachelor/Master thesis: Simulative image analysis (SIMBA) with AI and virtually generated training data
Fill develops fully automated production systems and supports the quality assurance of the produced components with automated inspection stations for detecting and sorting out defective parts. Great results have already been achieved with Convolutional Neural Networks (CNN). However, many applications lack the amount of training data.
The goal of this work is to use style transfer learning to generate training data for presence, type, and also surface inspections. Using a small number of real images and images from a simulation environment, a style context will be learned. This style transformation will then be used to automatically generate a large variation of training data with high information content in the simulation. The system shall be able to robustly train an AI (e.g. CNN) with a minimal number of real images, by aiding the simulative data.
In implementing this application, the project owner will gain extensive insight into machine vision processes and technologies with a focus on Deep Learning.
- Familiarization with the function of current learning approaches
- Annotating and labeling of training/testing data
- Development of a method for style transfer learning
- Evaluation of reliability and cost effectiveness
- Ongoing studies at a university of applied sciences/university
- Motivation & reliability
- Assumption of printing costs in case of very good or good results
- Support from a supervisor from the relevant specialist department
- Get to know Fill as a potential employer and contribute your own ideas and knowledge
- Great opportunity to supplement your theoretical knowledge with practical experience
- Very good working atmosphere in an award-winning family company
- Time frame: Start immediately
We look forward to receiving your application!
Application as Bachelor/Master thesis: Simulative image analysis (SIMBA) with AI and virtually generated training data.
Please fill in all fields, the mandatory fields are marked with *!